ai/openai: surface token usage on streams

Request stream_options.include_usage and return the final usage chunk as a Response with Usage set, so streaming callers can record usage.
ai/atlascloud: surface token usage on streams
2026-06-28 23:13:22 +01:00 · 2026-06-28 23:13:21 +01:00
54 changed files with 167 additions and 2372 deletions
@@ -33,8 +33,6 @@ jobs:
        run: go run ./internal/harness/universe
      - name: Agent-flow harness
        run: go run ./internal/harness/agent-flow
-      - name: Provider conformance (mock)
-        run: go run ./internal/harness/provider-conformance -providers mock
      - name: 0→hero run/chat/inspect reference scenario
        run: ./internal/harness/zero-to-hero-ci/run.sh

@@ -52,6 +50,17 @@ jobs:
        with:
          go-version: stable
          cache: true
+      - name: Agent provider conformance matrix
+        env:
+          GO_MICRO_AGENT_CONFORMANCE_LIVE: "1"
+          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
+          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
+          GROQ_API_KEY: ${{ secrets.GROQ_API_KEY }}
+          MISTRAL_API_KEY: ${{ secrets.MISTRAL_API_KEY }}
+          TOGETHER_API_KEY: ${{ secrets.TOGETHER_API_KEY }}
+          ATLASCLOUD_API_KEY: ${{ secrets.ATLASCLOUD_API_KEY }}
+        run: go test ./agent -run TestAgentProviderConformanceMatrix -count=1 -v
      - name: Provider conformance against live models
        env:
          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
@@ -18,7 +18,7 @@ help:
 	@echo "  make test-race     - Run tests with race detector"
 	@echo "  make test-coverage - Run tests with coverage"
 	@echo "  make lint          - Run linter"
-	@echo "  make harness       - Run deterministic getting-started and end-to-end harnesses"
+	@echo "  make harness       - Run deterministic end-to-end harnesses"
 	@echo "  make provider-conformance - Run harnesses against configured live providers"
 	@echo "  make fmt           - Format code"
 	@echo "  make install-tools - Install development tools"
@@ -42,14 +42,12 @@ test-coverage:
 	go tool cover -html=coverage.out -o coverage.html
 	@echo "Coverage report: coverage.html"

-# Run the documented getting-started contracts plus the deterministic
-# services → agents → workflows harnesses (mock LLM — no API key).
-# This mirrors the default CI path so local dogfooding catches scaffold,
-# run/chat/inspect, and 0→hero regressions before a PR is opened.
+# Run the end-to-end harnesses (deterministic, mock LLM — no API key).
+# The universe harness exits non-zero on assertion failure.
 harness:
-	go test ./cmd/micro/cli/new -run TestZeroToOneContract -count=1
-	./internal/harness/zero-to-hero-ci/run.sh
+	go run ./internal/harness/universe
 	go run ./internal/harness/agent-flow
+	go run ./internal/harness/plan-delegate # 0→hero: services + agents + flow + plan/delegate

 # Run the same harnesses against every configured live provider. Providers
 # without API keys are skipped; configured providers must pass.
@@ -139,35 +139,19 @@ func (a *agentImpl) String() string {
 }

 func (a *agentImpl) setup() {
-	a.setupWithToolHandler(nil)
-}
-
-func (a *agentImpl) setupWithToolHandler(handler ai.ToolHandler) {
 	var modelOpts []ai.Option
 	modelOpts = append(modelOpts, ai.WithAPIKey(a.opts.APIKey))
 	if a.opts.Model != "" {
 		modelOpts = append(modelOpts, ai.WithModel(a.opts.Model))
 	}

-	// Reuse the existing tools instance: its name map is populated by
-	// discoverTools, and rebuilding it here would orphan a base handler that
-	// already captured the old instance (breaking StreamAsk tool resolution).
-	if a.tools == nil {
-		a.tools = ai.NewTools(a.opts.Registry, ai.ToolClient(a.opts.Client))
-	}
-	if handler == nil {
-		handler = a.toolHandler()
-	}
-	modelOpts = append(modelOpts, ai.WithToolHandler(handler))
+	a.tools = ai.NewTools(a.opts.Registry, ai.ToolClient(a.opts.Client))
+	modelOpts = append(modelOpts, ai.WithToolHandler(a.toolHandler()))
 	a.model = ai.New(a.opts.Provider, modelOpts...)
 	if a.model != nil {
 		a.model = a.tracedModel(a.model)
 	}

-	if a.mem != nil {
-		return
-	}
-
 	// Memory is pluggable. Use the configured one, otherwise the default
 	// store-backed memory — except ephemeral sub-agents, which keep an
 	// isolated, non-persistent context.
@@ -241,18 +225,16 @@ func (a *agentImpl) ask(ctx context.Context, message, parentRunID string) (*Resp
 		a.setup()
 	}

-	return a.askLocked(ctx, uuid.New().String(), message, parentRunID, nil, true)
+	return a.askLocked(ctx, uuid.New().String(), message, parentRunID, nil)
 }

-func (a *agentImpl) askLocked(ctx context.Context, runID, message, parentRunID string, existing *flow.Run, addUserMessage bool) (*Response, error) {
+func (a *agentImpl) askLocked(ctx context.Context, runID, message, parentRunID string, existing *flow.Run) (*Response, error) {
 	toolList, err := a.discoverTools()
 	if err != nil {
 		return nil, fmt.Errorf("discover tools: %w", err)
 	}

-	if addUserMessage {
-		a.mem.Add("user", message)
-	}
+	a.mem.Add("user", message)
 	a.steps = 0
 	a.calls = map[string]int{}
 	a.pause = nil
@@ -113,14 +113,13 @@ func Builtins(opts ...Option) (tools []ai.Tool, handle func(name string, input m
 // prevents runaway recursion).
 func (a *agentImpl) toolHandler() ai.ToolHandler {
 	if a.ephemeral {
-		return a.toolTimeoutWrap(a.tools.Handler())
+		return a.tools.Handler()
 	}

 	// Innermost first: base, then guardrails (approve → loop → step →
 	// plan), then developer wrappers outermost. Wrapping reverses order,
 	// so the result runs plan → step → loop → approve → checkpoint → base.
 	h := a.baseHandler()
-	h = a.toolTimeoutWrap(h)
 	h = a.checkpointToolWrap(h)
 	h = a.approveWrap(h)
 	h = a.loopWrap(h)
@@ -149,21 +148,6 @@ func contextWrap(next ai.ToolHandler) ai.ToolHandler {
 	}
 }

-// toolTimeoutWrap gives each tool execution its own deadline while preserving
-// caller cancellation. Handlers still execute synchronously; tools that honor
-// context (custom tools, delegate RPC/A2A, and go-micro RPC clients) return
-// promptly with a bounded error result when the deadline expires.
-func (a *agentImpl) toolTimeoutWrap(next ai.ToolHandler) ai.ToolHandler {
-	return func(ctx context.Context, call ai.ToolCall) ai.ToolResult {
-		if a.opts.ToolTimeout <= 0 {
-			return next(ctx, call)
-		}
-		toolCtx, cancel := context.WithTimeout(ctx, a.opts.ToolTimeout)
-		defer cancel()
-		return next(toolCtx, call)
-	}
-}
-
 // baseHandler executes a tool call: a developer custom tool, the built-in
 // delegate, or an RPC to the service. It is the innermost handler.
 func (a *agentImpl) baseHandler() ai.ToolHandler {
@@ -335,9 +319,6 @@ func (a *agentImpl) handleDelegate(ctx context.Context, call ai.ToolCall) ai.Too
 		WithRegistry(a.opts.Registry),
 		WithClient(a.opts.Client),
 		WithStore(a.opts.Store),
-		ModelCallTimeout(a.opts.ModelTimeout),
-		ModelRetry(a.opts.ModelMaxAttempts, a.opts.ModelRetryBackoff),
-		ToolCallTimeout(a.opts.ToolTimeout),
 		TraceProvider(a.opts.TraceProvider),
 	)
 	// Record lineage so the sub-agent's tool calls carry this run as parent.
@@ -95,7 +95,7 @@ func (a *agentImpl) resume(ctx context.Context, runID string) (*Response, error)
 	if a.model == nil {
 		a.setup()
 	}
-	return a.askLocked(ctx, run.ID, message, parentID, &run, false)
+	return a.askLocked(ctx, run.ID, message, parentID, &run)
 }

 // ResumeInput resumes a checkpointed agent run that paused via the built-in
@@ -140,7 +140,7 @@ func (a *agentImpl) resumeInput(ctx context.Context, runID, input string) (*Resp
 	if a.model == nil {
 		a.setup()
 	}
-	return a.askLocked(ctx, run.ID, message, run.ParentID, &run, true)
+	return a.askLocked(ctx, run.ID, message, run.ParentID, &run)
 }

 func (a *agentImpl) pending(ctx context.Context) ([]flow.Run, error) {
@@ -102,72 +102,6 @@ func TestResumeFailedCheckpointDoesNotReplayCompletedTool(t *testing.T) {
 	}
 }

-func TestResumeFailedCheckpointDoesNotDuplicateCompactedMemory(t *testing.T) {
-	ctx := context.Background()
-	st := store.NewMemoryStore()
-	cp := flow.StoreCheckpoint(st, "memory-resume-agent")
-	failRetry := true
-	var sawRecall bool
-	fakeGen = func(ctx context.Context, opts ai.Options, req *ai.Request) (*ai.Response, error) {
-		for _, msg := range req.Messages {
-			if text, ok := msg.Content.(string); ok && strings.Contains(text, "alpha code is 42") {
-				sawRecall = true
-			}
-		}
-		if strings.Contains(req.Prompt, "use alpha code") && failRetry {
-			failRetry = false
-			return nil, errors.New("model connection dropped")
-		}
-		return &ai.Response{Reply: "ok"}, nil
-	}
-	defer func() { fakeGen = nil }()
-
-	a := newTestAgent(Name("memory-resume-agent"), WithStore(st), WithCheckpoint(cp), CompactMemory(4, 1), MemoryRecallLimit(2))
-	for _, msg := range []string{"alpha code is 42", "beta note", "gamma note"} {
-		if _, err := a.Ask(ctx, msg); err != nil {
-			t.Fatalf("Ask(%q): %v", msg, err)
-		}
-	}
-
-	_, err := a.Ask(ctx, "use alpha code now")
-	if err == nil {
-		t.Fatal("Ask succeeded, want simulated provider failure")
-	}
-	if got := countMemoryContent(a.mem.Messages(), "use alpha code now"); got != 1 {
-		t.Fatalf("failed Ask stored prompt %d times, want 1", got)
-	}
-
-	runs, err := Pending(ctx, a)
-	if err != nil {
-		t.Fatalf("Pending: %v", err)
-	}
-	if len(runs) != 1 {
-		t.Fatalf("Pending returned %d runs, want 1", len(runs))
-	}
-	if _, err := Resume(ctx, a, runs[0].ID); err != nil {
-		t.Fatalf("Resume: %v", err)
-	}
-	if got := countMemoryContent(a.mem.Messages(), "use alpha code now"); got != 1 {
-		t.Fatalf("resumed failed Ask stored prompt %d times, want no duplicate", got)
-	}
-	if !sawRecall {
-		t.Fatal("resume did not retrieve archived compacted memory")
-	}
-	if got := len(a.mem.Messages()); got > 4 {
-		t.Fatalf("compacted memory retained %d messages after resume, want <= 4", got)
-	}
-}
-
-func countMemoryContent(messages []ai.Message, needle string) int {
-	var count int
-	for _, msg := range messages {
-		if text, ok := msg.Content.(string); ok && strings.Contains(text, needle) {
-			count++
-		}
-	}
-	return count
-}
-
 func TestPendingReturnsUnfinishedAgentRuns(t *testing.T) {
 	ctx := context.Background()
 	cp := flow.StoreCheckpoint(store.NewStore(), "pending-agent")
@@ -243,51 +177,6 @@ func TestHumanInputPauseResumesSameRunWithInput(t *testing.T) {
 	}
 }

-func TestHumanInputResumeHonorsCanceledContextAndLeavesRunPending(t *testing.T) {
-	ctx := context.Background()
-	cp := flow.StoreCheckpoint(store.NewStore(), "input-cancel-agent")
-	fakeGen = func(ctx context.Context, opts ai.Options, req *ai.Request) (*ai.Response, error) {
-		if opts.ToolHandler != nil {
-			opts.ToolHandler(ctx, ai.ToolCall{ID: "input-1", Name: toolHumanInput, Input: map[string]any{"prompt": "Approve deploy?"}})
-		}
-		return &ai.Response{Reply: "waiting"}, nil
-	}
-	defer func() { fakeGen = nil }()
-
-	a := newTestAgent(Name("input-cancel-agent"), WithCheckpoint(cp))
-	if _, err := a.Ask(ctx, "deploy the service"); err == nil {
-		t.Fatal("Ask succeeded, want input-required pause")
-	}
-	runs, err := Pending(ctx, a)
-	if err != nil {
-		t.Fatalf("Pending: %v", err)
-	}
-	if len(runs) != 1 {
-		t.Fatalf("Pending returned %d runs, want 1: %#v", len(runs), runs)
-	}
-
-	canceled, cancel := context.WithCancel(ctx)
-	cancel()
-	if _, err := ResumeInput(canceled, a, runs[0].ID, "yes"); !errors.Is(err, context.Canceled) {
-		t.Fatalf("ResumeInput canceled err = %v, want context.Canceled", err)
-	}
-
-	loaded, ok, err := cp.Load(ctx, runs[0].ID)
-	if err != nil || !ok {
-		t.Fatalf("Load paused run ok=%v err=%v", ok, err)
-	}
-	if loaded.Status != "paused" || loaded.State.Stage != agentInputStep {
-		t.Fatalf("run status/stage after canceled resume = %s/%s, want paused/%s", loaded.Status, loaded.State.Stage, agentInputStep)
-	}
-	var pause inputPause
-	if err := loaded.State.Scan(&pause); err != nil {
-		t.Fatalf("Scan pause after canceled resume: %v", err)
-	}
-	if pause.OriginalMessage != "deploy the service" || pause.Prompt != "Approve deploy?" {
-		t.Fatalf("pause after canceled resume = %#v", pause)
-	}
-}
-
 func TestApprovalDenialPausesCheckpointedRunAndResumeContinues(t *testing.T) {
 	ctx := context.Background()
 	cp := flow.StoreCheckpoint(store.NewStore(), "approval-agent")
@@ -33,29 +33,14 @@ func TestAgentProviderConformanceMatrix(t *testing.T) {
 		{name: "together", key: "TOGETHER_API_KEY", model: "GO_MICRO_CONFORMANCE_TOGETHER_MODEL", live: true},
 	}

-	selected := selectedConformanceProviders(os.Getenv("GO_MICRO_AGENT_CONFORMANCE_PROVIDERS"))
 	for _, provider := range providers {
 		provider := provider
-		if len(selected) > 0 && !selected[provider.name] {
-			continue
-		}
 		t.Run(provider.name, func(t *testing.T) {
 			runAgentConformanceScenario(t, provider)
 		})
 	}
 }

-func selectedConformanceProviders(csv string) map[string]bool {
-	out := map[string]bool{}
-	for _, part := range strings.Split(csv, ",") {
-		part = strings.TrimSpace(part)
-		if part != "" {
-			out[part] = true
-		}
-	}
-	return out
-}
-
 func runAgentConformanceScenario(t *testing.T, provider conformanceProvider) {
 	t.Helper()
 	if provider.live {
@@ -56,10 +56,6 @@ type Options struct {
 	// ModelRetryBackoff is the base delay between transient provider failures
 	// (grows exponentially per attempt when retries are enabled).
 	ModelRetryBackoff time.Duration
-	// ToolTimeout bounds each tool execution (0 disables). The timeout is
-	// applied before custom tools, delegate, and service RPC calls so context
-	// deadlines propagate consistently through the agent loop.
-	ToolTimeout time.Duration

 	// Memory is the agent's conversation memory. Nil = the default
 	// store-backed memory (durable across restarts).
@@ -95,11 +91,6 @@ type Options struct {
 	// and tool calls. Nil disables instrumentation.
 	TraceProvider trace.TracerProvider

-	// TraceInputs controls whether agent observability records include raw
-	// user messages. It is false by default so spans and persisted run
-	// timelines carry correlation and shape without leaking prompts.
-	TraceInputs bool
-
 	// tools are developer-registered custom tools (see WithTool).
 	tools []customTool
 	// wrappers are developer-registered tool-execution wrappers
@@ -116,7 +107,6 @@ func newOptions(opts ...Option) Options {
 		ModelTimeout:      30 * time.Second,
 		ModelMaxAttempts:  1, // retries opt-in via ModelRetry (see field doc)
 		ModelRetryBackoff: 100 * time.Millisecond,
-		ToolTimeout:       30 * time.Second,
 		// On by default and lenient: identical repeated calls are a
 		// no-progress loop, never useful. Set LoopLimit(0) to disable.
 		LoopLimit: 3,
@@ -209,14 +199,6 @@ func ModelCallTimeout(d time.Duration) Option {
 	return func(o *Options) { o.ModelTimeout = d }
 }

-// ToolCallTimeout sets the timeout for each tool execution. It bounds custom
-// tools, built-in delegate calls, and service RPC tools with the same context
-// deadline so mid-run cancellation and slow tools produce safe error results
-// instead of unbounded agent runs. Set 0 to disable.
-func ToolCallTimeout(d time.Duration) Option {
-	return func(o *Options) { o.ToolTimeout = d }
-}
-
 // ModelRetry sets the provider retry budget and backoff for transient failures.
 func ModelRetry(maxAttempts int, backoff time.Duration) Option {
 	return func(o *Options) {
@@ -313,11 +295,3 @@ func WithTool(name, description string, properties map[string]any, handler ToolF
 func TraceProvider(tp trace.TracerProvider) Option {
 	return func(o *Options) { o.TraceProvider = tp }
 }
-
-// TraceInputs opts in to recording raw user messages on agent run events.
-// By default inputs are redacted from OpenTelemetry spans and persisted run
-// timelines; use this only when the observability backend is approved to store
-// prompt content.
-func TraceInputs(enabled bool) Option {
-	return func(o *Options) { o.TraceInputs = enabled }
-}
@@ -31,33 +31,27 @@ const (
 	AttrInputTokens    = "agent.tokens.input"
 	AttrOutputTokens   = "agent.tokens.output"
 	AttrTotalTokens    = "agent.tokens.total"
-	AttrAttempt        = "agent.model.attempt"
-	AttrMaxAttempts    = "agent.model.max_attempts"
 	AttrToolName       = "agent.tool.name"
 	AttrDelegate       = "agent.delegate"
 	AttrGuardrailBlock = "agent.guardrail.block"
 	AttrRefusal        = "agent.refusal"
-	AttrInputChars     = "agent.input.chars"
 )

 type RunEvent struct {
-	Time        time.Time `json:"time"`
-	RunID       string    `json:"run_id"`
-	ParentID    string    `json:"parent_id,omitempty"`
-	TraceID     string    `json:"trace_id,omitempty"`
-	SpanID      string    `json:"span_id,omitempty"`
-	Agent       string    `json:"agent"`
-	Kind        string    `json:"kind"`
-	Name        string    `json:"name,omitempty"`
-	Provider    string    `json:"provider,omitempty"`
-	Model       string    `json:"model,omitempty"`
-	Attempt     int       `json:"attempt,omitempty"`
-	MaxAttempts int       `json:"max_attempts,omitempty"`
-	LatencyMS   int64     `json:"latency_ms,omitempty"`
-	Tokens      Usage     `json:"tokens,omitempty"`
-	Refused     string    `json:"refused,omitempty"`
-	Error       string    `json:"error,omitempty"`
-	InputChars  int       `json:"input_chars,omitempty"`
+	Time      time.Time `json:"time"`
+	RunID     string    `json:"run_id"`
+	ParentID  string    `json:"parent_id,omitempty"`
+	TraceID   string    `json:"trace_id,omitempty"`
+	SpanID    string    `json:"span_id,omitempty"`
+	Agent     string    `json:"agent"`
+	Kind      string    `json:"kind"`
+	Name      string    `json:"name,omitempty"`
+	Provider  string    `json:"provider,omitempty"`
+	Model     string    `json:"model,omitempty"`
+	LatencyMS int64     `json:"latency_ms,omitempty"`
+	Tokens    Usage     `json:"tokens,omitempty"`
+	Refused   string    `json:"refused,omitempty"`
+	Error     string    `json:"error,omitempty"`
 }

 type Usage = ai.Usage
@@ -100,13 +94,9 @@ func (a *agentImpl) tracer() trace.Tracer {
 func (a *agentImpl) startRun(ctx context.Context, message string) (context.Context, func(error)) {
 	info, _ := ai.RunInfoFrom(ctx)
 	start := time.Now()
-	runEvent := RunEvent{Time: start, RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "run", InputChars: len(message)}
-	if a.opts.TraceInputs {
-		runEvent.Name = message
-	}

 	if a.opts.TraceProvider == nil {
-		a.recordRunEvent(runEvent)
+		a.recordRunEvent(RunEvent{Time: start, RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "run", Name: message})
 		return ctx, func(err error) {
 			latency := time.Since(start).Milliseconds()
 			if err != nil {
@@ -119,7 +109,7 @@ func (a *agentImpl) startRun(ctx context.Context, message string) (context.Conte

 	ctx, span := a.tracer().Start(ctx, spanNameRun, trace.WithSpanKind(trace.SpanKindInternal), trace.WithAttributes(
 		attribute.String(AttrRunID, info.RunID), attribute.String(AttrParentRunID, info.ParentID), attribute.String(AttrAgentName, info.Agent)))
-	a.recordSpanEvent(span, runEvent)
+	a.recordSpanEvent(span, RunEvent{Time: start, RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "run", Name: message})
 	return ctx, func(err error) {
 		latency := time.Since(start).Milliseconds()
 		span.SetAttributes(attribute.Int64(AttrLatencyMS, latency))
@@ -154,7 +144,7 @@ func (m *tracedModel) Generate(ctx context.Context, req *ai.Request, opts ...ai.
 		if resp != nil {
 			usage = resp.Usage
 		}
-		e := RunEvent{Time: time.Now(), RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "model", Provider: provider, Model: model, Attempt: info.Attempt, MaxAttempts: info.MaxAttempts, LatencyMS: dur, Tokens: usage}
+		e := RunEvent{Time: time.Now(), RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "model", Provider: provider, Model: model, LatencyMS: dur, Tokens: usage}
 		if err != nil {
 			e.Error = err.Error()
 		}
@@ -172,12 +162,6 @@ func (m *tracedModel) Generate(ctx context.Context, req *ai.Request, opts ...ai.
 	resp, err := m.Model.Generate(ctx, req, opts...)
 	dur := time.Since(start).Milliseconds()
 	attrs := []attribute.KeyValue{attribute.Int64(AttrLatencyMS, dur)}
-	if info.Attempt > 0 {
-		attrs = append(attrs, attribute.Int(AttrAttempt, info.Attempt))
-	}
-	if info.MaxAttempts > 0 {
-		attrs = append(attrs, attribute.Int(AttrMaxAttempts, info.MaxAttempts))
-	}
 	usage := ai.Usage{}
 	if resp != nil {
 		usage = resp.Usage
@@ -191,7 +175,7 @@ func (m *tracedModel) Generate(ctx context.Context, req *ai.Request, opts ...ai.
 		span.SetStatus(codes.Ok, "")
 	}
 	span.End()
-	e := RunEvent{Time: time.Now(), RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "model", Provider: provider, Model: model, Attempt: info.Attempt, MaxAttempts: info.MaxAttempts, LatencyMS: dur, Tokens: usage}
+	e := RunEvent{Time: time.Now(), RunID: info.RunID, ParentID: info.ParentID, Agent: info.Agent, Kind: "model", Provider: provider, Model: model, LatencyMS: dur, Tokens: usage}
 	if err != nil {
 		e.Error = err.Error()
 	}
@@ -290,18 +274,9 @@ func runEventAttributes(e RunEvent) []attribute.KeyValue {
 	if e.Model != "" {
 		attrs = append(attrs, attribute.String(AttrModel, e.Model))
 	}
-	if e.Attempt > 0 {
-		attrs = append(attrs, attribute.Int(AttrAttempt, e.Attempt))
-	}
-	if e.MaxAttempts > 0 {
-		attrs = append(attrs, attribute.Int(AttrMaxAttempts, e.MaxAttempts))
-	}
 	if e.LatencyMS > 0 {
 		attrs = append(attrs, attribute.Int64(AttrLatencyMS, e.LatencyMS))
 	}
-	if e.InputChars > 0 {
-		attrs = append(attrs, attribute.Int(AttrInputChars, e.InputChars))
-	}
 	attrs = appendUsage(attrs, e.Tokens)
 	if e.Refused != "" {
 		attrs = append(attrs, attribute.Bool(AttrGuardrailBlock, true), attribute.String(AttrRefusal, e.Refused))
@@ -3,7 +3,6 @@ package agent
 import (
 	"context"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"strings"
 	"testing"
@@ -12,13 +11,10 @@ import (
 	"go-micro.dev/v6/ai"
 	"go-micro.dev/v6/store"
 	"go.opentelemetry.io/otel/attribute"
-	"go.opentelemetry.io/otel/codes"
 	"go.opentelemetry.io/otel/sdk/trace"
 	"go.opentelemetry.io/otel/sdk/trace/tracetest"
 )

-const codesError = codes.Error
-
 type otelTestModel struct{ opts ai.Options }

 func (m *otelTestModel) Init(opts ...ai.Option) error {
@@ -93,9 +89,6 @@ func TestAgentOpenTelemetrySpans(t *testing.T) {
 		if attrs[AttrRunID] != runID || attrs[AttrAgentName] != "runner" {
 			t.Fatalf("%s missing run correlation attributes: %#v", s.Name(), attrs)
 		}
-		if s.Name() == spanNameModelCall && (attrs[AttrAttempt] != "1" || attrs[AttrMaxAttempts] != "1") {
-			t.Fatalf("model span missing attempt attributes: %#v", attrs)
-		}
 	}
 	keys, err := store.Scope(st, "agent", "runner").List(store.ListPrefix("runs/"))
 	if err != nil {
@@ -132,146 +125,6 @@ func TestAgentOpenTelemetrySpans(t *testing.T) {
 	}
 }

-func TestAgentRunObservabilityRedactsInputByDefault(t *testing.T) {
-	secret := "deploy production with token sk-secret"
-	exp := tracetest.NewInMemoryExporter()
-	tp := trace.NewTracerProvider(trace.WithSyncer(exp))
-	st := store.NewMemoryStore()
-	a := New(Name("redactor"), Provider("oteltest"), WithStore(st), TraceProvider(tp))
-	if _, err := a.Ask(context.Background(), secret); err != nil {
-		t.Fatal(err)
-	}
-
-	spans := exp.GetSpans().Snapshots()
-	var sawInputChars bool
-	for _, s := range spans {
-		for _, event := range s.Events() {
-			attrs := spanAttributes(event.Attributes)
-			if attrs["agent.event.name"] == secret {
-				t.Fatalf("span event leaked raw input: %#v", event)
-			}
-			if attrs[AttrInputChars] == fmt.Sprint(len(secret)) {
-				sawInputChars = true
-			}
-		}
-	}
-	if !sawInputChars {
-		t.Fatal("run event missing redacted input length attribute")
-	}
-
-	summaries, err := ListRunSummaries(st, "redactor")
-	if err != nil {
-		t.Fatal(err)
-	}
-	events, err := LoadRunEvents(st, "redactor", summaries[0].RunID)
-	if err != nil {
-		t.Fatal(err)
-	}
-	for _, event := range events {
-		if event.Name == secret {
-			t.Fatalf("persisted run event leaked raw input: %#v", event)
-		}
-		if event.Kind == "run" && event.InputChars != len(secret) {
-			t.Fatalf("run event InputChars = %d, want %d", event.InputChars, len(secret))
-		}
-	}
-}
-
-func TestAgentTraceInputsOptInRecordsInput(t *testing.T) {
-	message := "operator-approved diagnostic prompt"
-	st := store.NewMemoryStore()
-	a := New(Name("input-opt-in"), Provider("oteltest"), WithStore(st), TraceInputs(true))
-	if _, err := a.Ask(context.Background(), message); err != nil {
-		t.Fatal(err)
-	}
-
-	summaries, err := ListRunSummaries(st, "input-opt-in")
-	if err != nil {
-		t.Fatal(err)
-	}
-	events, err := LoadRunEvents(st, "input-opt-in", summaries[0].RunID)
-	if err != nil {
-		t.Fatal(err)
-	}
-	for _, event := range events {
-		if event.Kind == "run" && event.Name == message {
-			return
-		}
-	}
-	t.Fatalf("opt-in run event did not record message: %#v", events)
-}
-
-type failingOtelModel struct{ opts ai.Options }
-
-func (m *failingOtelModel) Init(opts ...ai.Option) error {
-	for _, o := range opts {
-		o(&m.opts)
-	}
-	return nil
-}
-func (m *failingOtelModel) Options() ai.Options { return m.opts }
-func (m *failingOtelModel) String() string      { return "otelfail" }
-func (m *failingOtelModel) Stream(context.Context, *ai.Request, ...ai.GenerateOption) (ai.Stream, error) {
-	return nil, nil
-}
-func (m *failingOtelModel) Generate(context.Context, *ai.Request, ...ai.GenerateOption) (*ai.Response, error) {
-	return nil, errors.New("provider exploded")
-}
-
-func init() {
-	ai.Register("otelfail", func(opts ...ai.Option) ai.Model { return &failingOtelModel{opts: ai.NewOptions(opts...)} })
-}
-
-func TestAgentOpenTelemetrySpansModelFailure(t *testing.T) {
-	exp := tracetest.NewInMemoryExporter()
-	tp := trace.NewTracerProvider(trace.WithSyncer(exp))
-	st := store.NewMemoryStore()
-	a := New(Name("failing-runner"), Provider("otelfail"), WithStore(st), TraceProvider(tp))
-	if _, err := a.Ask(context.Background(), "hello"); err == nil {
-		t.Fatal("Ask succeeded, want provider error")
-	}
-
-	spans := exp.GetSpans().Snapshots()
-	var sawRunError, sawModelError bool
-	for _, s := range spans {
-		attrs := spanAttributes(s.Attributes())
-		switch s.Name() {
-		case spanNameRun:
-			if attrs[AttrAgentName] == "failing-runner" && s.Status().Code == codesError {
-				sawRunError = true
-			}
-		case spanNameModelCall:
-			if attrs[AttrAgentName] == "failing-runner" && attrs[AttrAttempt] == "1" && s.Status().Code == codesError {
-				sawModelError = true
-			}
-		}
-	}
-	if !sawRunError || !sawModelError {
-		t.Fatalf("missing error spans: run=%v model=%v spans=%d", sawRunError, sawModelError, len(spans))
-	}
-
-	summaries, err := ListRunSummaries(st, "failing-runner")
-	if err != nil {
-		t.Fatal(err)
-	}
-	if len(summaries) != 1 || summaries[0].Status != "error" || summaries[0].LastError == "" {
-		t.Fatalf("unexpected failure summary: %#v", summaries)
-	}
-	events, err := LoadRunEvents(st, "failing-runner", summaries[0].RunID)
-	if err != nil {
-		t.Fatal(err)
-	}
-	var sawModelEvent bool
-	for _, event := range events {
-		if event.Kind == "model" && event.Attempt == 1 && event.MaxAttempts == 1 && event.Error != "" {
-			sawModelEvent = true
-		}
-	}
-	if !sawModelEvent {
-		t.Fatalf("missing failed model event with attempt metadata: %#v", events)
-	}
-}
-
 func spanEventHasRunInfo(events []trace.Event, name, runID, agentName string) bool {
 	for _, event := range events {
 		if event.Name != name {
@@ -102,30 +102,3 @@ func TestCanceledAskContextSkipsToolExecution(t *testing.T) {
 		t.Fatalf("plan persisted after canceled tool context: %q", plan)
 	}
 }
-
-func TestToolCallTimeoutPropagatesDeadlineToCustomTool(t *testing.T) {
-	var sawDeadline bool
-	a := newTestAgent(
-		Name("tool-timeout"),
-		ToolCallTimeout(10*time.Millisecond),
-		WithTool("slow", "slow tool", nil, func(ctx context.Context, input map[string]any) (string, error) {
-			if _, ok := ctx.Deadline(); ok {
-				sawDeadline = true
-			}
-			<-ctx.Done()
-			return "", ctx.Err()
-		}),
-	)
-
-	start := time.Now()
-	content := toolContent(a.toolHandler(), "slow", nil)
-	if !sawDeadline {
-		t.Fatal("custom tool did not receive a deadline")
-	}
-	if !strings.Contains(content, context.DeadlineExceeded.Error()) {
-		t.Fatalf("tool result = %q, want deadline exceeded", content)
-	}
-	if elapsed := time.Since(start); elapsed > 200*time.Millisecond {
-		t.Fatalf("tool call took %s, want bounded timeout", elapsed)
-	}
-}
@@ -1,237 +0,0 @@
-package agent
-
-import (
-	"context"
-	"encoding/json"
-	"errors"
-	"io"
-	"strings"
-	"sync"
-
-	"github.com/google/uuid"
-
-	"go-micro.dev/v6/ai"
-)
-
-// StreamEventType identifies an event emitted by a tool-aware agent stream.
-type StreamEventType string
-
-const (
-	// StreamEventToolStart is emitted immediately before a tool call runs.
-	StreamEventToolStart StreamEventType = "tool_start"
-	// StreamEventToolEnd is emitted after a tool call returns or is refused.
-	StreamEventToolEnd StreamEventType = "tool_end"
-	// StreamEventToken carries a chunk of the final answer.
-	StreamEventToken StreamEventType = "token"
-	// StreamEventDone carries the completed agent response.
-	StreamEventDone StreamEventType = "done"
-)
-
-// StreamEvent is one event from StreamAsk.
-type StreamEvent struct {
-	Type     StreamEventType
-	Token    string
-	ToolCall ai.ToolCall
-	Result   ai.ToolResult
-	Response *Response
-}
-
-// AgentStream is a stream of tool execution events followed by final-answer chunks.
-type AgentStream interface {
-	Recv() (*StreamEvent, error)
-	Close() error
-}
-
-// StreamAsk runs an agent Ask turn with tool start/end events and streams the final answer.
-// It is additive for callers that hold the public Agent interface; concrete agents also
-// expose the same method directly.
-func StreamAsk(ctx context.Context, ag Agent, message string) (AgentStream, error) {
-	streamer, ok := ag.(interface {
-		StreamAsk(context.Context, string) (AgentStream, error)
-	})
-	if !ok {
-		return nil, errors.New("agent: StreamAsk unsupported by implementation")
-	}
-	return streamer.StreamAsk(ctx, message)
-}
-
-// ResumeStreamAsk resumes a checkpointed agent run and emits the same event
-// shape as StreamAsk. Completed runs are streamed from the persisted response;
-// unfinished runs continue from their checkpoint and emit tool events for any
-// work that still needs to run. Tool calls already recorded as done in the
-// checkpoint are reused by the agent checkpoint wrapper and are not re-executed.
-func ResumeStreamAsk(ctx context.Context, ag Agent, runID string) (AgentStream, error) {
-	a, ok := ag.(*agentImpl)
-	if !ok {
-		return nil, errors.New("agent: ResumeStreamAsk unsupported by implementation")
-	}
-	return a.resumeStreamAsk(ctx, runID)
-}
-
-// StreamAsk runs tools like Ask, emits ToolStart/ToolEnd events as they execute,
-// then emits chunks of the final answer followed by a Done event.
-func (a *agentImpl) StreamAsk(ctx context.Context, message string) (AgentStream, error) {
-	events := make(chan *StreamEvent, 16)
-	done := make(chan struct{})
-	s := &agentStream{events: events, done: done}
-
-	go func() {
-		defer close(events)
-		defer close(done)
-		resp, err := a.askWithStreamEvents(ctx, message, events)
-		if err != nil {
-			s.setErr(err)
-			return
-		}
-		for _, tok := range splitStreamTokens(resp.Reply) {
-			if !sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToken, Token: tok}) {
-				return
-			}
-		}
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventDone, Response: resp})
-	}()
-	return s, nil
-}
-
-func (a *agentImpl) resumeStreamAsk(ctx context.Context, runID string) (AgentStream, error) {
-	events := make(chan *StreamEvent, 16)
-	done := make(chan struct{})
-	s := &agentStream{events: events, done: done}
-
-	go func() {
-		defer close(events)
-		defer close(done)
-		resp, err := a.resumeWithStreamEvents(ctx, runID, events)
-		if err != nil {
-			s.setErr(err)
-			return
-		}
-		for _, tok := range splitStreamTokens(resp.Reply) {
-			if !sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToken, Token: tok}) {
-				return
-			}
-		}
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventDone, Response: resp})
-	}()
-	return s, nil
-}
-
-func (a *agentImpl) askWithStreamEvents(ctx context.Context, message string, events chan<- *StreamEvent) (*Response, error) {
-	a.mu.Lock()
-	defer a.mu.Unlock()
-
-	if a.tools == nil {
-		a.tools = ai.NewTools(a.opts.Registry, ai.ToolClient(a.opts.Client))
-	}
-	base := a.toolHandler()
-	handler := func(ctx context.Context, call ai.ToolCall) ai.ToolResult {
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToolStart, ToolCall: call})
-		result := base(ctx, call)
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToolEnd, ToolCall: call, Result: result})
-		return result
-	}
-	a.setupWithToolHandler(handler)
-	defer a.setupWithToolHandler(nil)
-	return a.askLocked(ctx, uuid.New().String(), message, a.parentRunID, nil, true)
-}
-
-func (a *agentImpl) resumeWithStreamEvents(ctx context.Context, runID string, events chan<- *StreamEvent) (*Response, error) {
-	if a.opts.Checkpoint == nil {
-		return nil, errors.New("agent: ResumeStreamAsk requires a checkpoint")
-	}
-	run, ok, err := a.opts.Checkpoint.Load(ctx, runID)
-	if err != nil {
-		return nil, err
-	}
-	if !ok {
-		return nil, errors.New("agent: checkpointed run not found")
-	}
-	if run.Status == "done" {
-		var resp Response
-		if err := json.Unmarshal(run.State.Data, &resp); err != nil {
-			return nil, err
-		}
-		return &resp, nil
-	}
-
-	a.mu.Lock()
-	defer a.mu.Unlock()
-	if a.tools == nil {
-		a.tools = ai.NewTools(a.opts.Registry, ai.ToolClient(a.opts.Client))
-	}
-	base := a.toolHandler()
-	handler := func(ctx context.Context, call ai.ToolCall) ai.ToolResult {
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToolStart, ToolCall: call})
-		result := base(ctx, call)
-		_ = sendStreamEvent(ctx, events, &StreamEvent{Type: StreamEventToolEnd, ToolCall: call, Result: result})
-		return result
-	}
-	a.setupWithToolHandler(handler)
-	defer a.setupWithToolHandler(nil)
-	if run.Status == "paused" {
-		if run.State.Stage == agentInputStep {
-			return nil, errors.New("agent: checkpointed run is input-required; resume with ResumeInput")
-		}
-		run.Status = "running"
-		run.State.Stage = agentAskStep
-	}
-	return a.askLocked(ctx, run.ID, string(run.State.Data), run.ParentID, &run, false)
-}
-
-type agentStream struct {
-	events <-chan *StreamEvent
-	done   <-chan struct{}
-	mu     sync.Mutex
-	err    error
-}
-
-func (s *agentStream) Recv() (*StreamEvent, error) {
-	ev, ok := <-s.events
-	if ok {
-		return ev, nil
-	}
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	if s.err != nil {
-		return nil, s.err
-	}
-	return nil, io.EOF
-}
-
-func (s *agentStream) Close() error {
-	<-s.done
-	return nil
-}
-
-func (s *agentStream) setErr(err error) {
-	s.mu.Lock()
-	defer s.mu.Unlock()
-	s.err = err
-}
-
-func sendStreamEvent(ctx context.Context, events chan<- *StreamEvent, ev *StreamEvent) bool {
-	select {
-	case events <- ev:
-		return true
-	case <-ctx.Done():
-		return false
-	}
-}
-
-func splitStreamTokens(reply string) []string {
-	if reply == "" {
-		return nil
-	}
-	parts := strings.Fields(reply)
-	if len(parts) == 0 {
-		return []string{reply}
-	}
-	out := make([]string, 0, len(parts))
-	for i, part := range parts {
-		if i > 0 {
-			part = " " + part
-		}
-		out = append(out, part)
-	}
-	return out
-}
@@ -1,175 +0,0 @@
-package agent
-
-import (
-	"context"
-	"errors"
-	"io"
-	"testing"
-
-	"go-micro.dev/v6/ai"
-	"go-micro.dev/v6/flow"
-	"go-micro.dev/v6/store"
-)
-
-func TestStreamAskEmitsToolEventsAndFinalTokens(t *testing.T) {
-	calls := 0
-	fakeGen = func(ctx context.Context, opts ai.Options, req *ai.Request) (*ai.Response, error) {
-		if opts.ToolHandler == nil {
-			t.Fatal("StreamAsk must configure a tool handler")
-		}
-		calls++
-		result := opts.ToolHandler(ctx, ai.ToolCall{ID: "call-1", Name: "echo", Input: map[string]any{"text": "hello"}})
-		return &ai.Response{
-			Reply:     "planning",
-			Answer:    "final answer",
-			ToolCalls: []ai.ToolCall{{ID: "call-1", Name: "echo", Input: map[string]any{"text": "hello"}, Result: result.Content}},
-		}, nil
-	}
-	defer func() { fakeGen = nil }()
-
-	a := newTestAgent(Name("streamer"), WithTool("echo", "echo text", nil, func(ctx context.Context, input map[string]any) (string, error) {
-		return input["text"].(string), nil
-	}))
-	stream, err := a.StreamAsk(context.Background(), "say hello")
-	if err != nil {
-		t.Fatalf("StreamAsk: %v", err)
-	}
-
-	var types []StreamEventType
-	var tokens string
-	var done *Response
-	for {
-		event, err := stream.Recv()
-		if errors.Is(err, io.EOF) {
-			break
-		}
-		if err != nil {
-			t.Fatalf("Recv: %v", err)
-		}
-		types = append(types, event.Type)
-		if event.Type == StreamEventToken {
-			tokens += event.Token
-		}
-		if event.Type == StreamEventDone {
-			done = event.Response
-		}
-	}
-
-	want := []StreamEventType{StreamEventToolStart, StreamEventToolEnd, StreamEventToken, StreamEventToken, StreamEventToken, StreamEventDone}
-	if len(types) != len(want) {
-		t.Fatalf("event types = %v, want %v", types, want)
-	}
-	for i := range want {
-		if types[i] != want[i] {
-			t.Fatalf("event types = %v, want %v", types, want)
-		}
-	}
-	if tokens != "planning final answer" {
-		t.Fatalf("tokens = %q", tokens)
-	}
-	if done == nil || done.Reply != "planning\n\nfinal answer" {
-		t.Fatalf("done response = %#v", done)
-	}
-	if calls != 1 {
-		t.Fatalf("Generate calls = %d, want 1", calls)
-	}
-}
-
-func TestStreamAskHelperRejectsUnsupportedAgent(t *testing.T) {
-	_, err := StreamAsk(context.Background(), unsupportedAgent{}, "hello")
-	if err == nil {
-		t.Fatal("StreamAsk helper should reject unsupported implementations")
-	}
-}
-
-func TestResumeStreamAskDoesNotReplayCompletedTool(t *testing.T) {
-	ctx := context.Background()
-	cp := flow.StoreCheckpoint(store.NewStore(), "stream-resume-agent")
-	toolRuns := 0
-	first := true
-	fakeGen = func(ctx context.Context, opts ai.Options, req *ai.Request) (*ai.Response, error) {
-		if opts.ToolHandler != nil {
-			res := opts.ToolHandler(ctx, ai.ToolCall{ID: "call-1", Name: "charge", Input: map[string]any{"order": "42"}})
-			if res.Content != "charged" {
-				t.Fatalf("tool result = %q, want charged", res.Content)
-			}
-		}
-		if first {
-			first = false
-			return nil, errors.New("stream disconnected after tool")
-		}
-		return &ai.Response{Reply: "finished from streamed checkpoint"}, nil
-	}
-	defer func() { fakeGen = nil }()
-
-	a := newTestAgent(Name("stream-resume-agent"), WithCheckpoint(cp),
-		WithTool("charge", "charge once", nil, func(context.Context, map[string]any) (string, error) {
-			toolRuns++
-			return "charged", nil
-		}))
-	stream, err := a.StreamAsk(ctx, "charge order 42")
-	if err != nil {
-		t.Fatalf("StreamAsk: %v", err)
-	}
-	for {
-		_, err := stream.Recv()
-		if errors.Is(err, io.EOF) {
-			break
-		}
-		if err != nil {
-			break
-		}
-	}
-	if toolRuns != 1 {
-		t.Fatalf("tool executions after failed StreamAsk = %d, want 1", toolRuns)
-	}
-	runs, err := Pending(ctx, a)
-	if err != nil {
-		t.Fatalf("Pending: %v", err)
-	}
-	if len(runs) != 1 {
-		t.Fatalf("Pending returned %d runs, want 1", len(runs))
-	}
-
-	resumed, err := ResumeStreamAsk(ctx, a, runs[0].ID)
-	if err != nil {
-		t.Fatalf("ResumeStreamAsk: %v", err)
-	}
-	var toolEvents int
-	var done *Response
-	for {
-		event, err := resumed.Recv()
-		if errors.Is(err, io.EOF) {
-			break
-		}
-		if err != nil {
-			t.Fatalf("resumed Recv: %v", err)
-		}
-		if event.Type == StreamEventToolStart || event.Type == StreamEventToolEnd {
-			toolEvents++
-		}
-		if event.Type == StreamEventDone {
-			done = event.Response
-		}
-	}
-	if toolRuns != 1 {
-		t.Fatalf("tool executions after ResumeStreamAsk = %d, want completed tool was not replayed", toolRuns)
-	}
-	if toolEvents != 2 {
-		t.Fatalf("resumed tool events = %d, want start/end for replayed checkpoint result", toolEvents)
-	}
-	if done == nil || done.Reply != "finished from streamed checkpoint" || done.RunID != runs[0].ID {
-		t.Fatalf("done response = %#v", done)
-	}
-}
-
-type unsupportedAgent struct{}
-
-func (unsupportedAgent) Name() string                                      { return "unsupported" }
-func (unsupportedAgent) Init(...Option)                                    {}
-func (unsupportedAgent) Options() Options                                  { return Options{} }
-func (unsupportedAgent) Ask(context.Context, string) (*Response, error)    { return nil, nil }
-func (unsupportedAgent) Stream(context.Context, string) (ai.Stream, error) { return nil, nil }
-func (unsupportedAgent) Run() error                                        { return nil }
-func (unsupportedAgent) Stop() error                                       { return nil }
-func (unsupportedAgent) String() string                                    { return "unsupported" }
@@ -43,7 +43,6 @@ func init() {
 	ai.RegisterVideo("atlascloud", func(opts ...ai.Option) ai.VideoModel {
 		return NewProvider(opts...)
 	})
-	ai.RegisterStream("atlascloud")
 }

 // Provider implements the ai.Model interface for Atlas Cloud.
@@ -2,11 +2,7 @@ package atlascloud

 import (
 	"context"
-	"encoding/json"
 	"errors"
-	"io"
-	"net/http"
-	"net/http/httptest"
 	"testing"

 	"go-micro.dev/v6/ai"
@@ -85,58 +81,16 @@ func TestProvider_Generate_NoAPIKey(t *testing.T) {
 	}
 }

-func TestProvider_Stream(t *testing.T) {
-	var sawStream, sawIncludeUsage bool
-	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/v1/chat/completions" {
-			t.Errorf("path = %s, want /v1/chat/completions", r.URL.Path)
-		}
-		var body map[string]any
-		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
-			t.Fatalf("decode request: %v", err)
-		}
-		sawStream, _ = body["stream"].(bool)
-		if so, ok := body["stream_options"].(map[string]any); ok {
-			sawIncludeUsage, _ = so["include_usage"].(bool)
-		}
-		w.Header().Set("Content-Type", "text/event-stream")
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"hel\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"lo\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: {\"choices\":[],\"usage\":{\"prompt_tokens\":7,\"completion_tokens\":2,\"total_tokens\":9}}\n\n"))
-		_, _ = w.Write([]byte("data: [DONE]\n\n"))
-	}))
-	defer ts.Close()
+func TestProvider_Stream_NotImplemented(t *testing.T) {
+	p := NewProvider()

-	p := NewProvider(ai.WithAPIKey("test-key"), ai.WithBaseURL(ts.URL))
-	stream, err := p.Stream(context.Background(), &ai.Request{Prompt: "Hello"})
-	if err != nil {
-		t.Fatalf("Stream returned error: %v", err)
-	}
-	defer stream.Close()
-	if !sawStream {
-		t.Fatal("stream request did not set stream=true")
-	}
-	if !sawIncludeUsage {
-		t.Fatal("stream request did not set stream_options.include_usage=true")
+	req := &ai.Request{
+		Prompt: "Hello",
 	}

-	first, err := stream.Recv()
-	if err != nil || first.Reply != "hel" {
-		t.Fatalf("first chunk = %#v, %v; want hel", first, err)
-	}
-	second, err := stream.Recv()
-	if err != nil || second.Reply != "lo" {
-		t.Fatalf("second chunk = %#v, %v; want lo", second, err)
-	}
-	usage, err := stream.Recv()
-	if err != nil {
-		t.Fatalf("usage chunk error: %v", err)
-	}
-	if usage.Usage.TotalTokens != 9 || usage.Usage.InputTokens != 7 || usage.Usage.OutputTokens != 2 {
-		t.Fatalf("usage = %#v; want input=7 output=2 total=9", usage.Usage)
-	}
-	if _, err := stream.Recv(); !errors.Is(err, io.EOF) {
-		t.Fatalf("final error = %v, want EOF", err)
+	_, err := p.Stream(context.Background(), req)
+	if !errors.Is(err, ai.ErrStreamingUnsupported) {
+		t.Fatalf("Stream error = %v, want ErrStreamingUnsupported", err)
 	}
 }

@@ -34,7 +34,7 @@ func TestRegisteredProviders(t *testing.T) {
 	}

 	got = ai.RegisteredProviders("stream")
-	want = []string{"atlascloud", "groq", "mistral", "openai", "together"}
+	want = []string{"openai"}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("RegisteredProviders(stream) = %#v, want %#v", got, want)
 	}
@@ -44,12 +44,12 @@ func TestCapabilityRows(t *testing.T) {
 	got := ai.CapabilityRows()
 	want := []ai.CapabilityRow{
 		{Provider: "anthropic", Capabilities: ai.Capabilities{Model: true}},
-		{Provider: "atlascloud", Capabilities: ai.Capabilities{Model: true, Image: true, Video: true, Stream: true}},
+		{Provider: "atlascloud", Capabilities: ai.Capabilities{Model: true, Image: true, Video: true}},
 		{Provider: "gemini", Capabilities: ai.Capabilities{Model: true}},
-		{Provider: "groq", Capabilities: ai.Capabilities{Model: true, Stream: true}},
-		{Provider: "mistral", Capabilities: ai.Capabilities{Model: true, Stream: true}},
+		{Provider: "groq", Capabilities: ai.Capabilities{Model: true}},
+		{Provider: "mistral", Capabilities: ai.Capabilities{Model: true}},
 		{Provider: "openai", Capabilities: ai.Capabilities{Model: true, Image: true, Stream: true}},
-		{Provider: "together", Capabilities: ai.Capabilities{Model: true, Stream: true}},
+		{Provider: "together", Capabilities: ai.Capabilities{Model: true}},
 	}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("CapabilityRows() = %#v, want %#v", got, want)
@@ -72,7 +72,7 @@ func TestCapabilityMatrix(t *testing.T) {
 	if caps := ai.ProviderCapabilities("openai"); caps != (ai.Capabilities{Model: true, Image: true, Stream: true}) {
 		t.Fatalf("ProviderCapabilities(openai) = %#v", caps)
 	}
-	if caps := ai.ProviderCapabilities("atlascloud"); caps != (ai.Capabilities{Model: true, Image: true, Video: true, Stream: true}) {
+	if caps := ai.ProviderCapabilities("atlascloud"); caps != (ai.Capabilities{Model: true, Image: true, Video: true}) {
 		t.Fatalf("ProviderCapabilities(atlascloud) = %#v", caps)
 	}
 	if caps := ai.ProviderCapabilities("missing"); caps != (ai.Capabilities{}) {
@@ -88,7 +88,7 @@ func TestRegisterStream(t *testing.T) {
 	}

 	got := ai.RegisteredProviders("stream")
-	want := []string{"atlascloud", "groq", "mistral", "openai", "test-stream", "together"}
+	want := []string{"openai", "test-stream"}
 	if !reflect.DeepEqual(got, want) {
 		t.Fatalf("RegisteredProviders(stream) = %#v, want %#v", got, want)
 	}
@@ -22,14 +22,12 @@ import (
 	"strings"

 	"go-micro.dev/v6/ai"
-	"go-micro.dev/v6/ai/internal/openaiapi"
 )

 func init() {
 	ai.Register("groq", func(opts ...ai.Option) ai.Model {
 		return NewProvider(opts...)
 	})
-	ai.RegisterStream("groq")
 }

 type Provider struct {
@@ -121,7 +119,7 @@ func (p *Provider) Generate(ctx context.Context, req *ai.Request, opts ...ai.Gen
 }

 func (p *Provider) Stream(ctx context.Context, req *ai.Request, opts ...ai.GenerateOption) (ai.Stream, error) {
-	return openaiapi.Stream(ctx, p.opts, req, "/v1/chat/completions")
+	return nil, fmt.Errorf("%w: groq provider", ai.ErrStreamingUnsupported)
 }

 func (p *Provider) callAPI(ctx context.Context, req map[string]any) (*ai.Response, map[string]any, error) {
@@ -2,11 +2,7 @@ package groq

 import (
 	"context"
-	"encoding/json"
 	"errors"
-	"io"
-	"net/http"
-	"net/http/httptest"
 	"testing"

 	"go-micro.dev/v6/ai"
@@ -44,44 +40,9 @@ func TestProvider_Generate_NoAPIKey(t *testing.T) {
 	}
 }

-func TestProvider_Stream(t *testing.T) {
-	var sawStream bool
-	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/v1/chat/completions" {
-			t.Fatalf("path = %s, want /v1/chat/completions", r.URL.Path)
-		}
-		var body map[string]any
-		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
-			t.Fatalf("decode request: %v", err)
-		}
-		sawStream, _ = body["stream"].(bool)
-		w.Header().Set("Content-Type", "text/event-stream")
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"hel\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"lo\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: [DONE]\n\n"))
-	}))
-	defer ts.Close()
-
-	p := NewProvider(ai.WithAPIKey("test-key"), ai.WithBaseURL(ts.URL))
-	stream, err := p.Stream(context.Background(), &ai.Request{Prompt: "Hello"})
-	if err != nil {
-		t.Fatalf("Stream returned error: %v", err)
-	}
-	defer stream.Close()
-	if !sawStream {
-		t.Fatal("stream request did not set stream=true")
-	}
-
-	first, err := stream.Recv()
-	if err != nil || first.Reply != "hel" {
-		t.Fatalf("first chunk = %#v, %v; want hel", first, err)
-	}
-	second, err := stream.Recv()
-	if err != nil || second.Reply != "lo" {
-		t.Fatalf("second chunk = %#v, %v; want lo", second, err)
-	}
-	if _, err := stream.Recv(); !errors.Is(err, io.EOF) {
-		t.Fatalf("final error = %v, want EOF", err)
+func TestProvider_Stream_NotImplemented(t *testing.T) {
+	if _, err := NewProvider().Stream(context.Background(), &ai.Request{Prompt: "hi"}); !errors.Is(err, ai.ErrStreamingUnsupported) {
+		t.Fatalf("Stream error = %v, want ErrStreamingUnsupported", err)
 	}
 }

@@ -1,117 +0,0 @@
-package openaiapi
-
-import (
-	"bufio"
-	"bytes"
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"net/http"
-	"strings"
-
-	"go-micro.dev/v6/ai"
-)
-
-// Stream opens an OpenAI-compatible chat completions SSE stream.
-func Stream(ctx context.Context, opts ai.Options, req *ai.Request, basePath string) (ai.Stream, error) {
-	messages := []map[string]any{{"role": "system", "content": req.SystemPrompt}}
-	for _, m := range req.Messages {
-		messages = append(messages, map[string]any{"role": m.Role, "content": m.Content})
-	}
-	if req.Prompt != "" {
-		messages = append(messages, map[string]any{"role": "user", "content": req.Prompt})
-	}
-	apiReq := map[string]any{
-		"model":          opts.Model,
-		"messages":       messages,
-		"stream":         true,
-		"stream_options": map[string]any{"include_usage": true},
-	}
-	if opts.MaxTokens > 0 {
-		apiReq["max_tokens"] = opts.MaxTokens
-	}
-	reqBody, err := json.Marshal(apiReq)
-	if err != nil {
-		return nil, fmt.Errorf("failed to marshal stream request: %w", err)
-	}
-	apiURL := strings.TrimRight(opts.BaseURL, "/") + basePath
-	httpReq, err := http.NewRequestWithContext(ctx, http.MethodPost, apiURL, bytes.NewReader(reqBody))
-	if err != nil {
-		return nil, fmt.Errorf("failed to create stream request: %w", err)
-	}
-	httpReq.Header.Set("Content-Type", "application/json")
-	httpReq.Header.Set("Accept", "text/event-stream")
-	httpReq.Header.Set("Authorization", "Bearer "+opts.APIKey)
-
-	httpResp, err := http.DefaultClient.Do(httpReq)
-	if err != nil {
-		return nil, fmt.Errorf("stream API request failed: %w", err)
-	}
-	if httpResp.StatusCode != http.StatusOK {
-		defer httpResp.Body.Close()
-		respBody, _ := io.ReadAll(httpResp.Body)
-		return nil, fmt.Errorf("stream API error (%s): %s", httpResp.Status, string(respBody))
-	}
-	return &StreamReader{body: httpResp.Body, scanner: bufio.NewScanner(httpResp.Body)}, nil
-}
-
-// StreamReader reads OpenAI-compatible server-sent event chunks.
-type StreamReader struct {
-	body    io.ReadCloser
-	scanner *bufio.Scanner
-	closed  bool
-}
-
-func (s *StreamReader) Recv() (*ai.Response, error) {
-	for s.scanner.Scan() {
-		line := strings.TrimSpace(s.scanner.Text())
-		if line == "" || strings.HasPrefix(line, ":") {
-			continue
-		}
-		if !strings.HasPrefix(line, "data:") {
-			continue
-		}
-		data := strings.TrimSpace(strings.TrimPrefix(line, "data:"))
-		if data == "[DONE]" {
-			return nil, io.EOF
-		}
-		var chunk struct {
-			Choices []struct {
-				Delta struct {
-					Content string `json:"content"`
-				} `json:"delta"`
-			} `json:"choices"`
-			Usage *struct {
-				PromptTokens     int `json:"prompt_tokens"`
-				CompletionTokens int `json:"completion_tokens"`
-				TotalTokens      int `json:"total_tokens"`
-			} `json:"usage"`
-		}
-		if err := json.Unmarshal([]byte(data), &chunk); err != nil {
-			return nil, fmt.Errorf("failed to parse stream chunk: %w", err)
-		}
-		if len(chunk.Choices) > 0 && chunk.Choices[0].Delta.Content != "" {
-			return &ai.Response{Reply: chunk.Choices[0].Delta.Content}, nil
-		}
-		if chunk.Usage != nil {
-			return &ai.Response{Usage: ai.Usage{
-				InputTokens:  chunk.Usage.PromptTokens,
-				OutputTokens: chunk.Usage.CompletionTokens,
-				TotalTokens:  chunk.Usage.TotalTokens,
-			}}, nil
-		}
-	}
-	if err := s.scanner.Err(); err != nil {
-		return nil, err
-	}
-	return nil, io.EOF
-}
-
-func (s *StreamReader) Close() error {
-	if s.closed {
-		return nil
-	}
-	s.closed = true
-	return s.body.Close()
-}
@@ -22,14 +22,12 @@ import (
 	"strings"

 	"go-micro.dev/v6/ai"
-	"go-micro.dev/v6/ai/internal/openaiapi"
 )

 func init() {
 	ai.Register("mistral", func(opts ...ai.Option) ai.Model {
 		return NewProvider(opts...)
 	})
-	ai.RegisterStream("mistral")
 }

 type Provider struct {
@@ -121,7 +119,7 @@ func (p *Provider) Generate(ctx context.Context, req *ai.Request, opts ...ai.Gen
 }

 func (p *Provider) Stream(ctx context.Context, req *ai.Request, opts ...ai.GenerateOption) (ai.Stream, error) {
-	return openaiapi.Stream(ctx, p.opts, req, "/v1/chat/completions")
+	return nil, fmt.Errorf("%w: mistral provider", ai.ErrStreamingUnsupported)
 }

 func (p *Provider) callAPI(ctx context.Context, req map[string]any) (*ai.Response, map[string]any, error) {
@@ -2,11 +2,7 @@ package mistral

 import (
 	"context"
-	"encoding/json"
 	"errors"
-	"io"
-	"net/http"
-	"net/http/httptest"
 	"testing"

 	"go-micro.dev/v6/ai"
@@ -44,44 +40,9 @@ func TestProvider_Generate_NoAPIKey(t *testing.T) {
 	}
 }

-func TestProvider_Stream(t *testing.T) {
-	var sawStream bool
-	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/v1/chat/completions" {
-			t.Fatalf("path = %s, want /v1/chat/completions", r.URL.Path)
-		}
-		var body map[string]any
-		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
-			t.Fatalf("decode request: %v", err)
-		}
-		sawStream, _ = body["stream"].(bool)
-		w.Header().Set("Content-Type", "text/event-stream")
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"hel\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"lo\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: [DONE]\n\n"))
-	}))
-	defer ts.Close()
-
-	p := NewProvider(ai.WithAPIKey("test-key"), ai.WithBaseURL(ts.URL))
-	stream, err := p.Stream(context.Background(), &ai.Request{Prompt: "Hello"})
-	if err != nil {
-		t.Fatalf("Stream returned error: %v", err)
-	}
-	defer stream.Close()
-	if !sawStream {
-		t.Fatal("stream request did not set stream=true")
-	}
-
-	first, err := stream.Recv()
-	if err != nil || first.Reply != "hel" {
-		t.Fatalf("first chunk = %#v, %v; want hel", first, err)
-	}
-	second, err := stream.Recv()
-	if err != nil || second.Reply != "lo" {
-		t.Fatalf("second chunk = %#v, %v; want lo", second, err)
-	}
-	if _, err := stream.Recv(); !errors.Is(err, io.EOF) {
-		t.Fatalf("final error = %v, want EOF", err)
+func TestProvider_Stream_NotImplemented(t *testing.T) {
+	if _, err := NewProvider().Stream(context.Background(), &ai.Request{Prompt: "hi"}); !errors.Is(err, ai.ErrStreamingUnsupported) {
+		t.Fatalf("Stream error = %v, want ErrStreamingUnsupported", err)
 	}
 }

@@ -13,12 +13,6 @@ type StatusCoder interface {
 	StatusCode() int
 }

-// RetryAfterCoder is implemented by provider errors that expose a server
-// supplied retry delay, such as HTTP Retry-After on a 429/503 response.
-type RetryAfterCoder interface {
-	RetryAfter() time.Duration
-}
-
 // ErrorKind classifies provider-boundary failures into stable buckets callers
 // can inspect without parsing provider-specific error strings.
 type ErrorKind string
@@ -119,7 +113,16 @@ func GenerateWithRetry(ctx context.Context, m Model, req *Request, policy Genera
 		// Always back off between retries — exponential and capped — so an
 		// opt-in retry can never become a tight loop hammering the provider,
 		// even if Backoff was left at zero.
-		backoff := retryBackoff(err, attempt, policy.Backoff)
+		backoff := policy.Backoff
+		if backoff <= 0 {
+			backoff = 200 * time.Millisecond
+		}
+		if shift := attempt - 1; shift > 0 {
+			backoff <<= shift
+		}
+		if backoff > 30*time.Second {
+			backoff = 30 * time.Second
+		}
 		t := time.NewTimer(backoff)
 		select {
 		case <-ctx.Done():
@@ -133,30 +136,6 @@ func GenerateWithRetry(ctx context.Context, m Model, req *Request, policy Genera
 	return nil, &RetryError{Attempts: policy.MaxAttempts, Kind: ClassifyError(last), Err: last}
 }

-func retryBackoff(err error, attempt int, base time.Duration) time.Duration {
-	backoff := base
-	if backoff <= 0 {
-		backoff = 200 * time.Millisecond
-	}
-	if shift := attempt - 1; shift > 0 {
-		backoff <<= shift
-	}
-	if backoff > 30*time.Second {
-		backoff = 30 * time.Second
-	}
-
-	var retryAfter RetryAfterCoder
-	if errors.As(err, &retryAfter) {
-		if delay := retryAfter.RetryAfter(); delay > backoff {
-			backoff = delay
-		}
-	}
-	if backoff > 30*time.Second {
-		return 30 * time.Second
-	}
-	return backoff
-}
-
 // ClassifyError maps provider and context failures to stable operational kinds.
 func ClassifyError(err error) ErrorKind {
 	if err == nil {
@@ -139,14 +139,6 @@ type statusErr int
 func (e statusErr) Error() string   { return "provider status" }
 func (e statusErr) StatusCode() int { return int(e) }

-type retryAfterErr struct {
-	delay time.Duration
-}
-
-func (e retryAfterErr) Error() string             { return "rate limit exceeded" }
-func (e retryAfterErr) StatusCode() int           { return 429 }
-func (e retryAfterErr) RetryAfter() time.Duration { return e.delay }
-
 func TestClassifyErrorDistinguishesOperationalOutcomes(t *testing.T) {
 	tests := []struct {
 		name string
@@ -189,35 +181,3 @@ func TestGenerateWithRetryExposesRetryErrorKind(t *testing.T) {
 		t.Fatalf("retry error does not unwrap provider status: %v", err)
 	}
 }
-
-func TestGenerateWithRetryHonorsRetryAfterWhenLongerThanBackoff(t *testing.T) {
-	attempts := 0
-	model := retryModel{generate: func(context.Context, *Request, ...GenerateOption) (*Response, error) {
-		attempts++
-		if attempts == 1 {
-			return nil, retryAfterErr{delay: 25 * time.Millisecond}
-		}
-		return &Response{Reply: "ok"}, nil
-	}}
-
-	start := time.Now()
-	resp, err := GenerateWithRetry(context.Background(), model, &Request{Prompt: "hi"}, GeneratePolicy{
-		MaxAttempts: 2,
-		Backoff:     time.Millisecond,
-	})
-	if err != nil {
-		t.Fatalf("GenerateWithRetry returned error: %v", err)
-	}
-	if resp.Reply != "ok" {
-		t.Fatalf("reply = %q, want ok", resp.Reply)
-	}
-	if elapsed := time.Since(start); elapsed < 20*time.Millisecond {
-		t.Fatalf("retry delay = %s, want RetryAfter delay to dominate base backoff", elapsed)
-	}
-}
-
-func TestGenerateWithRetryCapsRetryAfter(t *testing.T) {
-	if got := retryBackoff(retryAfterErr{delay: time.Minute}, 1, time.Millisecond); got != 30*time.Second {
-		t.Fatalf("retryBackoff() = %s, want 30s cap", got)
-	}
-}
@@ -22,14 +22,12 @@ import (
 	"strings"

 	"go-micro.dev/v6/ai"
-	"go-micro.dev/v6/ai/internal/openaiapi"
 )

 func init() {
 	ai.Register("together", func(opts ...ai.Option) ai.Model {
 		return NewProvider(opts...)
 	})
-	ai.RegisterStream("together")
 }

 type Provider struct {
@@ -121,7 +119,7 @@ func (p *Provider) Generate(ctx context.Context, req *ai.Request, opts ...ai.Gen
 }

 func (p *Provider) Stream(ctx context.Context, req *ai.Request, opts ...ai.GenerateOption) (ai.Stream, error) {
-	return openaiapi.Stream(ctx, p.opts, req, "/v1/chat/completions")
+	return nil, fmt.Errorf("%w: together provider", ai.ErrStreamingUnsupported)
 }

 func (p *Provider) callAPI(ctx context.Context, req map[string]any) (*ai.Response, map[string]any, error) {
@@ -2,11 +2,7 @@ package together

 import (
 	"context"
-	"encoding/json"
 	"errors"
-	"io"
-	"net/http"
-	"net/http/httptest"
 	"testing"

 	"go-micro.dev/v6/ai"
@@ -44,44 +40,9 @@ func TestProvider_Generate_NoAPIKey(t *testing.T) {
 	}
 }

-func TestProvider_Stream(t *testing.T) {
-	var sawStream bool
-	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-		if r.URL.Path != "/v1/chat/completions" {
-			t.Fatalf("path = %s, want /v1/chat/completions", r.URL.Path)
-		}
-		var body map[string]any
-		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
-			t.Fatalf("decode request: %v", err)
-		}
-		sawStream, _ = body["stream"].(bool)
-		w.Header().Set("Content-Type", "text/event-stream")
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"hel\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: {\"choices\":[{\"delta\":{\"content\":\"lo\"}}]}\n\n"))
-		_, _ = w.Write([]byte("data: [DONE]\n\n"))
-	}))
-	defer ts.Close()
-
-	p := NewProvider(ai.WithAPIKey("test-key"), ai.WithBaseURL(ts.URL))
-	stream, err := p.Stream(context.Background(), &ai.Request{Prompt: "Hello"})
-	if err != nil {
-		t.Fatalf("Stream returned error: %v", err)
-	}
-	defer stream.Close()
-	if !sawStream {
-		t.Fatal("stream request did not set stream=true")
-	}
-
-	first, err := stream.Recv()
-	if err != nil || first.Reply != "hel" {
-		t.Fatalf("first chunk = %#v, %v; want hel", first, err)
-	}
-	second, err := stream.Recv()
-	if err != nil || second.Reply != "lo" {
-		t.Fatalf("second chunk = %#v, %v; want lo", second, err)
-	}
-	if _, err := stream.Recv(); !errors.Is(err, io.EOF) {
-		t.Fatalf("final error = %v, want EOF", err)
+func TestProvider_Stream_NotImplemented(t *testing.T) {
+	if _, err := NewProvider().Stream(context.Background(), &ai.Request{Prompt: "hi"}); !errors.Is(err, ai.ErrStreamingUnsupported) {
+		t.Fatalf("Stream error = %v, want ErrStreamingUnsupported", err)
 	}
 }

@@ -43,9 +43,6 @@ func Deploy(c *cli.Context) error {
 	}

 	target, remotePath := resolveDeployTarget(c, target, cfg)
-	if c.Bool("dry-run") {
-		return printDeployPlan(c, target, cfg, remotePath)
-	}

 	return deploySSH(c, target, cfg, remotePath)
 }
@@ -99,81 +96,6 @@ func showDeployTargets(cfg *config.Config) error {
 	return fmt.Errorf("%s", sb.String())
 }

-func printDeployPlan(c *cli.Context, target string, cfg *config.Config, remotePath string) error {
-	dir := c.Args().Get(1)
-	if dir == "" {
-		dir = "."
-	}
-
-	absDir, err := filepath.Abs(dir)
-	if err != nil {
-		return fmt.Errorf("failed to get absolute path: %w", err)
-	}
-
-	if cfg == nil {
-		cfg, _ = config.Load(absDir)
-	}
-	if remotePath == "" {
-		remotePath = defaultRemotePath
-	}
-
-	services, err := deployServices(absDir, cfg, c.String("service"))
-	if err != nil {
-		return err
-	}
-
-	fmt.Println()
-	fmt.Println("  \033[1mmicro deploy --dry-run\033[0m")
-	fmt.Println()
-	fmt.Printf("  Target      \033[36m%s\033[0m\n", target)
-	fmt.Printf("  Remote path %s\n", remotePath)
-	fmt.Printf("  Services    %s\n", strings.Join(services, ", "))
-	fmt.Println()
-	fmt.Println("  Plan:")
-	fmt.Println("    1. Build linux/amd64 service binaries")
-	fmt.Printf("    2. Copy binaries to %s/bin/\n", remotePath)
-	fmt.Println("    3. Enable and restart micro@<service> systemd units")
-	fmt.Println("    4. Check service health")
-	fmt.Println()
-	fmt.Println("  No SSH, rsync, systemd, or remote deployment was performed.")
-	return nil
-}
-
-func deployServices(absDir string, cfg *config.Config, filterService string) ([]string, error) {
-	if filterService != "" && cfg != nil {
-		found := false
-		for _, svc := range cfg.Services {
-			if svc.Name == filterService {
-				found = true
-				break
-			}
-		}
-		if !found && len(cfg.Services) > 0 {
-			return nil, fmt.Errorf("service '%s' not found in configuration", filterService)
-		}
-	}
-
-	if cfg != nil && len(cfg.Services) > 0 {
-		sorted, err := cfg.TopologicalSort()
-		if err != nil {
-			return nil, err
-		}
-		services := make([]string, 0, len(sorted))
-		for _, svc := range sorted {
-			if filterService == "" || svc.Name == filterService {
-				services = append(services, svc.Name)
-			}
-		}
-		return services, nil
-	}
-
-	services := []string{filepath.Base(absDir)}
-	if filterService != "" && filterService != services[0] {
-		return nil, fmt.Errorf("service '%s' not found (only '%s' available)", filterService, services[0])
-	}
-	return services, nil
-}
-
 func deploySSH(c *cli.Context, target string, cfg *config.Config, remotePath string) error {
 	dir := c.Args().Get(1)
 	if dir == "" {
@@ -199,10 +121,19 @@ func deploySSH(c *cli.Context, target string, cfg *config.Config, remotePath str
 	fmt.Println()
 	fmt.Printf("  Target     \033[36m%s\033[0m\n\n", target)

-	// Early validation: resolve services before SSH checks.
-	services, err := deployServices(absDir, cfg, c.String("service"))
-	if err != nil {
-		return err
+	// Early validation: Check if the requested service exists before SSH checks
+	filterService := c.String("service")
+	if filterService != "" && cfg != nil {
+		found := false
+		for _, svc := range cfg.Services {
+			if svc.Name == filterService {
+				found = true
+				break
+			}
+		}
+		if !found && len(cfg.Services) > 0 {
+			return fmt.Errorf("service '%s' not found in configuration", filterService)
+		}
 	}

 	// Step 1: Check SSH connectivity
@@ -222,6 +153,28 @@ func deploySSH(c *cli.Context, target string, cfg *config.Config, remotePath str
 	fmt.Println("\u2713")

 	// Step 3: Build binaries
+	var services []string
+	if cfg != nil && len(cfg.Services) > 0 {
+		sorted, err := cfg.TopologicalSort()
+		if err != nil {
+			return err
+		}
+		for _, svc := range sorted {
+			// If --service flag is provided, only include that service
+			if filterService == "" || svc.Name == filterService {
+				services = append(services, svc.Name)
+			}
+		}
+	} else {
+		// Single service project
+		services = []string{filepath.Base(absDir)}
+
+		// If --service flag was provided for a single-service project, validate it matches
+		if filterService != "" && filterService != services[0] {
+			return fmt.Errorf("service '%s' not found (only '%s' available)", filterService, services[0])
+		}
+	}
+
 	fmt.Printf("  Building binaries...       ")
 	if err := buildBinaries(absDir, cfg, c.Bool("build"), services); err != nil {
 		fmt.Println("\u2717")
@@ -540,10 +493,6 @@ The deploy process:
 				Name:  "service",
 				Usage: "Deploy only a specific service (for multi-service projects)",
 			},
-			&cli.BoolFlag{
-				Name:  "dry-run",
-				Usage: "Print the deployment plan without building, connecting, copying, or restarting services",
-			},
 		},
 	})
 }
@@ -17,7 +17,6 @@ func newDeployTestContext(t *testing.T, args ...string) *cli.Context {
 	set.String("ssh", "", "")
 	set.String("service", "", "")
 	set.Bool("build", false, "")
-	set.Bool("dry-run", false, "")
 	if err := set.Parse(args); err != nil {
 		t.Fatalf("parse flags: %v", err)
 	}
@@ -119,49 +118,3 @@ deploy prod
 		t.Fatalf("deploy target = %#v", prod)
 	}
 }
-
-func TestDeployDryRunPlansConfiguredTargetWithoutRemoteSideEffects(t *testing.T) {
-	dir := t.TempDir()
-	if err := os.WriteFile(dir+"/micro.mu", []byte(`service api
-    path ./api
-
-deploy prod
-    ssh deploy@prod.example.com
-    path /srv/micro
-`), 0644); err != nil {
-		t.Fatalf("write config: %v", err)
-	}
-
-	oldwd, err := os.Getwd()
-	if err != nil {
-		t.Fatalf("getwd: %v", err)
-	}
-	if err := os.Chdir(dir); err != nil {
-		t.Fatalf("chdir: %v", err)
-	}
-	t.Cleanup(func() {
-		if err := os.Chdir(oldwd); err != nil {
-			t.Errorf("restore cwd: %v", err)
-		}
-	})
-
-	ctx := newDeployTestContext(t, "--dry-run", "prod")
-	if err := Deploy(ctx); err != nil {
-		t.Fatalf("dry-run deploy: %v", err)
-	}
-}
-
-func TestDeployDryRunValidatesRequestedService(t *testing.T) {
-	ctx := newDeployTestContext(t, "--dry-run", "--service", "missing", "prod")
-	cfg := &config.Config{Services: map[string]*config.Service{
-		"api": {Name: "api", Path: "./api"},
-	}}
-
-	err := printDeployPlan(ctx, "deploy@prod.example.com", cfg, defaultRemotePath)
-	if err == nil {
-		t.Fatal("expected dry-run to validate service names")
-	}
-	if !strings.Contains(err.Error(), "service 'missing' not found in configuration") {
-		t.Fatalf("unexpected error: %v", err)
-	}
-}
@@ -1,169 +0,0 @@
-// Package inspect registers the 'micro inspect' CLI command.
-package inspect
-
-import (
-	"context"
-	"encoding/json"
-	"fmt"
-	"io"
-	"os"
-
-	"github.com/urfave/cli/v2"
-	goagent "go-micro.dev/v6/agent"
-	"go-micro.dev/v6/cmd"
-	aiflow "go-micro.dev/v6/flow"
-	"go-micro.dev/v6/store"
-)
-
-func init() {
-	cmd.Register(&cli.Command{
-		Name:  "inspect",
-		Usage: "Inspect recent agent and workflow activity",
-		Description: `Inspect is the CLI checkpoint in the local scaffold → run → chat → inspect loop.
-It reads durable local run history, so it works after the agent or flow has stopped.`,
-		Subcommands: []*cli.Command{
-			{
-				Name:      "agent",
-				Usage:     "Show recent recorded runs for an agent",
-				ArgsUsage: "[agent]",
-				Flags:     inspectAgentFlags(),
-				Action:    inspectAgent,
-			},
-			{
-				Name:      "flow",
-				Usage:     "Show durable run history for a flow",
-				ArgsUsage: "[flow]",
-				Flags:     inspectFlowFlags(),
-				Action:    inspectFlow,
-			},
-		},
-	})
-}
-
-func inspectAgentFlags() []cli.Flag {
-	return []cli.Flag{
-		&cli.BoolFlag{Name: "json", Usage: "Print run summaries as JSON for automation"},
-		&cli.StringFlag{Name: "status", Usage: "Only show runs with this status (running, done, error, refused)"},
-		&cli.StringFlag{Name: "trace", Usage: "Only show runs whose trace id matches this full id or prefix"},
-		&cli.IntFlag{Name: "limit", Usage: "Show the most recently updated N runs"},
-	}
-}
-
-func inspectFlowFlags() []cli.Flag {
-	return []cli.Flag{
-		&cli.BoolFlag{Name: "json", Usage: "Print durable run history as JSON for automation"},
-		&cli.BoolFlag{Name: "pending", Usage: "Only show runs that have not completed"},
-		&cli.StringFlag{Name: "status", Usage: "Only show runs with this status (running, done, failed)"},
-		&cli.IntFlag{Name: "limit", Usage: "Show the most recently updated N runs"},
-		&cli.StringFlag{Name: "stage", Usage: "Only show runs currently checkpointed at this stage"},
-	}
-}
-
-func inspectAgent(c *cli.Context) error {
-	name := c.Args().First()
-	if name == "" {
-		return fmt.Errorf("agent name required: micro inspect agent <name>")
-	}
-	opts := goagent.RunListOptions{Status: c.String("status"), TraceID: c.String("trace"), Limit: c.Int("limit")}
-	runs, err := goagent.ListRunSummariesWithOptions(store.DefaultStore, name, opts)
-	if err != nil {
-		return err
-	}
-	return writeAgentInspection(os.Stdout, name, runs, c.Bool("json"))
-}
-
-func writeAgentInspection(w io.Writer, name string, runs []goagent.RunSummary, asJSON bool) error {
-	if asJSON {
-		enc := json.NewEncoder(w)
-		enc.SetIndent("", "  ")
-		return enc.Encode(runs)
-	}
-	if len(runs) == 0 {
-		fmt.Fprintf(w, "  No agent runs recorded for %q. After chatting, try: micro inspect agent %s\n", name, name)
-		return nil
-	}
-	fmt.Fprintf(w, "  Agent %q runs\n", name)
-	for _, run := range runs {
-		fmt.Fprintf(w, "  %s  status=%s  events=%d  last=%s", run.RunID, run.Status, run.Events, run.LastKind)
-		if run.LastError != "" {
-			fmt.Fprintf(w, "  error=%q", run.LastError)
-		}
-		if run.TraceID != "" {
-			fmt.Fprintf(w, "  trace=%s", shortID(run.TraceID))
-		}
-		fmt.Fprintln(w)
-	}
-	return nil
-}
-
-func inspectFlow(c *cli.Context) error {
-	name := c.Args().First()
-	if name == "" {
-		return fmt.Errorf("flow name required: micro inspect flow <name>")
-	}
-	runs, err := aiflow.StoreCheckpoint(nil, name).List(context.Background())
-	if err != nil {
-		return err
-	}
-	runs = filterFlowInspection(runs, c.Bool("pending"), c.String("status"), c.String("stage"), c.Int("limit"))
-	return writeFlowInspection(os.Stdout, name, runs, c.Bool("json"), c.Bool("pending"))
-}
-
-func filterFlowInspection(runs []aiflow.Run, pending bool, status, stage string, limit int) []aiflow.Run {
-	filtered := make([]aiflow.Run, 0, len(runs))
-	for _, run := range runs {
-		if pending && run.Status == "done" {
-			continue
-		}
-		if status != "" && run.Status != status {
-			continue
-		}
-		if stage != "" && run.State.Stage != stage {
-			continue
-		}
-		filtered = append(filtered, run)
-	}
-	if limit > 0 && len(filtered) > limit {
-		return filtered[len(filtered)-limit:]
-	}
-	return filtered
-}
-
-func writeFlowInspection(w io.Writer, name string, runs []aiflow.Run, asJSON, pending bool) error {
-	if asJSON {
-		enc := json.NewEncoder(w)
-		enc.SetIndent("", "  ")
-		return enc.Encode(runs)
-	}
-	if len(runs) == 0 {
-		if pending {
-			fmt.Fprintf(w, "  No pending flow runs recorded for %q.\n", name)
-			return nil
-		}
-		fmt.Fprintf(w, "  No flow runs recorded for %q. After executing a durable flow, try: micro inspect flow %s\n", name, name)
-		return nil
-	}
-	fmt.Fprintf(w, "  Flow %q runs\n", name)
-	for _, run := range runs {
-		stage := run.State.Stage
-		if stage == "" {
-			stage = "-"
-		}
-		fmt.Fprintf(w, "  %s  status=%s  stage=%s  steps=%d", shortID(run.ID), run.Status, stage, len(run.Steps))
-		for _, step := range run.Steps {
-			if step.Error != "" {
-				fmt.Fprintf(w, "  error=%q", step.Error)
-				break
-			}
-		}
-		fmt.Fprintln(w)
-	}
-	return nil
-}
-
-func shortID(id string) string {
-	if len(id) <= 12 {
-		return id
-	}
-	return id[:12]
-}
@@ -1,64 +0,0 @@
-package inspect
-
-import (
-	"bytes"
-	"encoding/json"
-	"strings"
-	"testing"
-
-	goagent "go-micro.dev/v6/agent"
-	aiflow "go-micro.dev/v6/flow"
-)
-
-func TestWriteAgentInspectionIncludesActionableBreadcrumbs(t *testing.T) {
-	runs := []goagent.RunSummary{{RunID: "run-1", Status: "error", Events: 4, LastKind: "tool", LastError: "boom", TraceID: "1234567890abcdef"}}
-	var out bytes.Buffer
-	if err := writeAgentInspection(&out, "support", runs, false); err != nil {
-		t.Fatal(err)
-	}
-	got := out.String()
-	for _, want := range []string{"Agent \"support\" runs", "run-1", "status=error", "events=4", "last=tool", `error="boom"`, "trace=1234567890ab"} {
-		if !strings.Contains(got, want) {
-			t.Fatalf("output missing %q:\n%s", want, got)
-		}
-	}
-}
-
-func TestWriteAgentInspectionEmptyStateNamesInspectCommand(t *testing.T) {
-	var out bytes.Buffer
-	if err := writeAgentInspection(&out, "support", nil, false); err != nil {
-		t.Fatal(err)
-	}
-	if got := out.String(); !strings.Contains(got, "micro inspect agent support") {
-		t.Fatalf("empty state missing next step: %q", got)
-	}
-}
-
-func TestWriteFlowInspectionIncludesFailedStepBreadcrumb(t *testing.T) {
-	runs := []aiflow.Run{{ID: "1234567890abcdef", Status: "failed", State: aiflow.State{Stage: "charge"}, Steps: []aiflow.StepRecord{{Name: "charge", Status: "failed", Error: "card declined"}}}}
-	var out bytes.Buffer
-	if err := writeFlowInspection(&out, "checkout", runs, false, false); err != nil {
-		t.Fatal(err)
-	}
-	got := out.String()
-	for _, want := range []string{"Flow \"checkout\" runs", "1234567890ab", "status=failed", "stage=charge", "steps=1", `error="card declined"`} {
-		if !strings.Contains(got, want) {
-			t.Fatalf("output missing %q:\n%s", want, got)
-		}
-	}
-}
-
-func TestWriteFlowInspectionJSON(t *testing.T) {
-	runs := []aiflow.Run{{ID: "run-1", Flow: "checkout", Status: "done"}}
-	var out bytes.Buffer
-	if err := writeFlowInspection(&out, "checkout", runs, true, false); err != nil {
-		t.Fatal(err)
-	}
-	var got []aiflow.Run
-	if err := json.Unmarshal(out.Bytes(), &got); err != nil {
-		t.Fatalf("invalid JSON: %v\n%s", err, out.String())
-	}
-	if len(got) != 1 || got[0].ID != "run-1" || got[0].Status != "done" {
-		t.Fatalf("decoded runs = %+v", got)
-	}
-}
@@ -12,7 +12,6 @@ import (
 	_ "go-micro.dev/v6/cmd/micro/cli/build"
 	_ "go-micro.dev/v6/cmd/micro/cli/deploy"
 	_ "go-micro.dev/v6/cmd/micro/flow"
-	_ "go-micro.dev/v6/cmd/micro/inspect"
 	_ "go-micro.dev/v6/cmd/micro/mcp"
 	_ "go-micro.dev/v6/cmd/micro/resource"
 	_ "go-micro.dev/v6/cmd/micro/run"
@@ -19,7 +19,7 @@ func TestZeroToHeroCLIBoundaries(t *testing.T) {
 		}
 	}

-	for _, want := range []string{"run", "chat", "flow", "inspect", "deploy"} {
+	for _, want := range []string{"run", "chat", "flow"} {
 		if !commands[want] {
 			t.Fatalf("missing %q command", want)
 		}
@@ -27,24 +27,4 @@ func TestZeroToHeroCLIBoundaries(t *testing.T) {
 	if !subcommands["flow"]["runs"] {
 		t.Fatal("missing inspect boundary: flow runs")
 	}
-	if !subcommands["inspect"]["agent"] || !subcommands["inspect"]["flow"] {
-		t.Fatal("missing inspect boundary: inspect agent/flow")
-	}
-
-	var hasDeployDryRun bool
-	for _, command := range microcmd.DefaultCmd.App().Commands {
-		if command.Name != "deploy" {
-			continue
-		}
-		for _, flag := range command.Flags {
-			for _, name := range flag.Names() {
-				if name == "dry-run" {
-					hasDeployDryRun = true
-				}
-			}
-		}
-	}
-	if !hasDeployDryRun {
-		t.Fatal("missing deploy boundary: deploy --dry-run")
-	}
 }
@@ -19,11 +19,7 @@ func NewStream(opts ...Option) (Stream, error) {
 	for _, o := range opts {
 		o(&options)
 	}
-	st := options.Store
-	if st == nil {
-		st = store.NewMemoryStore()
-	}
-	return &mem{store: st}, nil
+	return &mem{store: store.NewMemoryStore()}, nil
 }

 type subscriber struct {
@@ -1,25 +1,11 @@
 package events

-import (
-	"time"
+import "time"

-	"go-micro.dev/v6/store"
-)
-
-type Options struct {
-	// Store persists published events for durability and replay. If nil, an
-	// in-memory store is used and events do not survive a restart.
-	Store store.Store
-}
+type Options struct{}

 type Option func(o *Options)

-// WithStore backs the stream with a durable store (e.g. the file store), so
-// published events persist and can be replayed across restarts.
-func WithStore(s store.Store) Option {
-	return func(o *Options) { o.Store = s }
-}
-
 type StoreOptions struct {
 	TTL    time.Duration
 	Backup Backup
@@ -80,11 +80,10 @@ A workflow as ordered, checkpointed steps that survives a crash and resumes wher
 - **Checkpoint** — each step is persisted; on `Resume`, completed steps are not re-run (no duplicate side effects)

 ### [support](./support/)
-A maintained 0-to-hero reference path in one runnable file:
- **scaffold** typed `customers`, `tickets`, and `notify` services
- **run/chat** with a support agent that uses those services as tools
- **inspect** the event-driven `intake` flow and approval gate
- **CI** keeps the deterministic mock-model journey runnable with `go test ./examples/support`
+A real-world support desk — the "zero to hero" shape in one runnable file:
+- **services** (`customers`, `tickets`, `notify`) become the agent's tools automatically
+- **flow** turns a `ticket.created` event into work for the agent (the event is the prompt)
+- **guardrail** — the agent triages freely but can't email a customer without passing the approval gate

 ## Coming Soon

@@ -34,11 +34,3 @@ calls that tool, Go Micro persists the run with status `paused` and stage
 `input-required`. Plain `agent.Resume` continues to support completed, failed,
 and approval-paused runs; input-required runs are resumed with
 `agent.ResumeInput` so the human response is explicit.
-
-## Cancellation and deadlines
-
-`ResumeInput` uses the caller's `context.Context` for checkpoint reads, writes,
-and the resumed model/tool turn. If the context is canceled or its deadline
-expires before the resume is committed, the call returns the context error and
-the checkpointed run remains `paused` at `input-required`; list it with
-`agent.Pending` and retry with a fresh context once the operator is ready.
@@ -1,25 +1,9 @@
-# Zero-to-hero support desk
+# Support desk

-A maintained 0-to-hero reference for the Go Micro lifecycle: scaffold a few
-typed services, run them in one process, let an agent chat with those services
-as tools, then inspect the durable flow that triggered the work. It is one
-runnable file and one CI smoke test, so the reference path stays honest as the
-framework evolves.
-
-## The path
-
-1. **Scaffold services** — `customers`, `tickets`, and `notify` are ordinary
-   typed Go Micro services. Their request/response structs and method comments
-   become the tool contract the agent sees.
-2. **Run the harness** — the example starts an in-memory registry, broker,
-   client, store, services, agent, and flow in one process; no external
-   dependencies or API key are required for the default run.
-3. **Chat through an agent** — the `support` agent receives the ticket event as
-   a prompt and calls service tools to look up the customer, triage the ticket,
-   and draft a reply.
-4. **Inspect the workflow** — the `intake` flow records the event-driven run and
-   prints the agent result, showing the service → agent → workflow lifecycle as
-   one runtime.
+A real-world agent built the Go Micro way: a few services, an agent that
+manages them, an event that triggers it, and a human-in-the-loop gate on the
+one action that touches a customer. It's the "zero to hero" shape in one
+runnable file.

 ## The scenario

@@ -60,12 +44,6 @@ agent, which:
 go run main.go            # mock model — deterministic, no API key
 ```

-The maintained check is the same deterministic path:
-
-```bash
-go test ./examples/support
-```
-
 Against a live model, the agent reasons about the ticket itself instead of
 following the script:

@@ -1,32 +1,9 @@
 package main

-import (
-	"os"
-	"strings"
-	"testing"
-)
+import "testing"

 func TestRunSupportMockSmoke(t *testing.T) {
 	if err := runSupport("mock"); err != nil {
 		t.Fatalf("support example failed: %v", err)
 	}
 }
-
-func TestZeroToHeroReadmeDocumentsLifecycle(t *testing.T) {
-	b, err := os.ReadFile("README.md")
-	if err != nil {
-		t.Fatalf("read README.md: %v", err)
-	}
-	doc := string(b)
-	for _, want := range []string{
-		"Scaffold services",
-		"Run the harness",
-		"Chat through an agent",
-		"Inspect the workflow",
-		"go test ./examples/support",
-	} {
-		if !strings.Contains(doc, want) {
-			t.Fatalf("README.md missing zero-to-hero step %q", want)
-		}
-	}
-}
@@ -140,11 +140,8 @@ func (g *Gateway) Handler() http.Handler {
 	// Per-agent card (served at the agent's url and at its well-known path).
 	mux.HandleFunc("GET /agents/{name}", g.handleCard)
 	mux.HandleFunc("GET /agents/{name}/.well-known/agent.json", g.handleCard)
-	mux.HandleFunc("GET /agents/{name}/skills/{skill}", g.handleSkillCard)
-	mux.HandleFunc("GET /agents/{name}/skills/{skill}/.well-known/agent.json", g.handleSkillCard)
 	// Per-agent JSON-RPC endpoint.
 	mux.HandleFunc("POST /agents/{name}", g.handleRPC)
-	mux.HandleFunc("POST /agents/{name}/skills/{skill}", g.handleSkillRPC)
 	// Top-level well-known: serve the single agent's card if there's
 	// exactly one, otherwise point to the directory.
 	mux.HandleFunc("GET /.well-known/agent.json", g.handleWellKnown)
@@ -341,17 +338,23 @@ func Card(name, url, description string, services []string) AgentCard {
 			description = "Go Micro agent"
 		}
 	}
-	skills := skillsFromServices(services)
 	return AgentCard{
-		Name:               name,
-		Description:        description,
-		URL:                url,
-		Version:            "1.0.0",
-		ProtocolVersion:    protocolVersion,
-		Capabilities:       Capabilities{Streaming: true, PushNotifications: true},
+		Name:            name,
+		Description:     description,
+		URL:             url,
+		Version:         "1.0.0",
+		ProtocolVersion: protocolVersion,
+		Capabilities:    Capabilities{Streaming: true, PushNotifications: true},
+		// The agent converses over a single Chat endpoint; advertise that
+		// as one skill, tagged with the services it manages.
 		DefaultInputModes:  []string{"text/plain"},
 		DefaultOutputModes: []string{"text/plain"},
-		Skills:             skills,
+		Skills: []Skill{{
+			ID:          "chat",
+			Name:        "Chat",
+			Description: "Converse with the agent to operate its services.",
+			Tags:        services,
+		}},
 	}
 }

@@ -368,19 +371,6 @@ func (g *Gateway) lookupCard(name string) (AgentCard, bool) {
 	return g.card(name, meta), true
 }

-func (g *Gateway) lookupSkillCard(name, skillID string) (AgentCard, Skill, bool) {
-	card, ok := g.lookupCard(name)
-	if !ok {
-		return AgentCard{}, Skill{}, false
-	}
-	for _, skill := range card.Skills {
-		if skill.ID == skillID {
-			return card, skill, true
-		}
-	}
-	return AgentCard{}, Skill{}, false
-}
-
 // ---------------------------------------------------------------------------
 // HTTP handlers
 // ---------------------------------------------------------------------------
@@ -403,17 +393,6 @@ func (g *Gateway) handleCard(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, http.StatusOK, card)
 }

-func (g *Gateway) handleSkillCard(w http.ResponseWriter, r *http.Request) {
-	card, skill, ok := g.lookupSkillCard(r.PathValue("name"), r.PathValue("skill"))
-	if !ok {
-		http.NotFound(w, r)
-		return
-	}
-	card.URL = g.opts.BaseURL + "/agents/" + r.PathValue("name") + "/skills/" + skill.ID
-	card.Skills = []Skill{skill}
-	writeJSON(w, http.StatusOK, card)
-}
-
 func (g *Gateway) handleWellKnown(w http.ResponseWriter, r *http.Request) {
 	cards, err := g.cards()
 	if err != nil {
@@ -442,18 +421,6 @@ func (g *Gateway) handleRPC(w http.ResponseWriter, r *http.Request) {
 	})
 }

-func (g *Gateway) handleSkillRPC(w http.ResponseWriter, r *http.Request) {
-	name := r.PathValue("name")
-	_, skill, ok := g.lookupSkillCard(name, r.PathValue("skill"))
-	if !ok {
-		writeRPC(w, nil, nil, &rpcError{Code: errInvalidParams, Message: "unknown agent skill: " + name + "/" + r.PathValue("skill")})
-		return
-	}
-	g.disp.serve(w, r, func(ctx context.Context, text string) (string, error) {
-		return g.callAgent(ctx, name, skillPrompt(skill, text))
-	})
-}
-
 // dispatcher handles A2A JSON-RPC requests against an Invoke function and
 // retains recent tasks for tasks/get. It is shared by the gateway (one
 // per registry) and embedded agents (one per agent).
@@ -886,68 +853,6 @@ func (d *dispatcher) deliverPush(taskID string, task *Task) {
 	}
 }

-func skillsFromServices(services []string) []Skill {
-	if len(services) == 0 {
-		return []Skill{{ID: "chat", Name: "Chat", Description: "Converse with the agent to operate its services."}}
-	}
-	seen := map[string]bool{}
-	var skills []Skill
-	for _, service := range services {
-		service = strings.TrimSpace(service)
-		if service == "" {
-			continue
-		}
-		id := skillID(service)
-		if id == "" || seen[id] {
-			continue
-		}
-		seen[id] = true
-		skills = append(skills, Skill{
-			ID:          id,
-			Name:        skillName(service),
-			Description: fmt.Sprintf("Operate the %s service through this agent.", service),
-			Tags:        []string{service},
-		})
-	}
-	if len(skills) == 0 {
-		return []Skill{{ID: "chat", Name: "Chat", Description: "Converse with the agent to operate its services."}}
-	}
-	return skills
-}
-
-func skillID(service string) string {
-	service = strings.ToLower(strings.TrimSpace(service))
-	var b strings.Builder
-	dash := false
-	for _, r := range service {
-		if (r >= 'a' && r <= 'z') || (r >= '0' && r <= '9') {
-			b.WriteRune(r)
-			dash = false
-			continue
-		}
-		if !dash && b.Len() > 0 {
-			b.WriteByte('-')
-			dash = true
-		}
-	}
-	return strings.Trim(b.String(), "-")
-}
-
-func skillName(service string) string {
-	parts := strings.FieldsFunc(service, func(r rune) bool { return r == '-' || r == '_' || r == '.' || r == '/' || r == ' ' })
-	for i, part := range parts {
-		if part == "" {
-			continue
-		}
-		parts[i] = strings.ToUpper(part[:1]) + part[1:]
-	}
-	return strings.Join(parts, " ")
-}
-
-func skillPrompt(skill Skill, text string) string {
-	return fmt.Sprintf("Use the %q skill (%s) for this request.\n\n%s", skill.Name, skill.ID, text)
-}
-
 func textOf(parts []Part) string {
 	var b strings.Builder
 	for _, p := range parts {
@@ -51,7 +51,7 @@ func newGatewayWithAgent(t *testing.T) (*httptest.Server, func()) {
 		server.Name("echo"),
 		server.Address("127.0.0.1:0"),
 		server.Registry(reg),
-		server.Metadata(map[string]string{"type": "agent", "services": "task,project"}),
+		server.Metadata(map[string]string{"type": "agent", "services": ""}),
 	)
 	if err := pb.RegisterAgentHandler(srv, echoAgent{}); err != nil {
 		t.Fatalf("register agent handler: %v", err)
@@ -88,40 +88,8 @@ func TestAgentCardFromRegistry(t *testing.T) {
 	if card.URL != "http://gw/agents/echo" {
 		t.Errorf("card url = %q", card.URL)
 	}
-	if card.ProtocolVersion == "" {
-		t.Errorf("card missing protocolVersion: %+v", card)
-	}
-	if got := skillIDs(card.Skills); strings.Join(got, ",") != "task,project" {
-		t.Errorf("skill IDs = %v, want [task project]", got)
-	}
-}
-
-func TestSkillEndpointServesFocusedCardAndRoutesRPC(t *testing.T) {
-	ts, cleanup := newGatewayWithAgent(t)
-	defer cleanup()
-
-	resp, err := http.Get(ts.URL + "/agents/echo/skills/task/.well-known/agent.json")
-	if err != nil {
-		t.Fatalf("get skill card: %v", err)
-	}
-	defer resp.Body.Close()
-	if resp.StatusCode != http.StatusOK {
-		t.Fatalf("skill card status = %d", resp.StatusCode)
-	}
-	var card AgentCard
-	if err := json.NewDecoder(resp.Body).Decode(&card); err != nil {
-		t.Fatalf("decode skill card: %v", err)
-	}
-	if card.URL != "http://gw/agents/echo/skills/task" || len(card.Skills) != 1 || card.Skills[0].ID != "task" {
-		t.Fatalf("skill card = %+v, want task-only card at skill URL", card)
-	}
-
-	task := rpcTask(t, ts.URL+"/agents/echo/skills/task", `{
-		"jsonrpc":"2.0","id":1,"method":"message/send",
-		"params":{"message":{"role":"user","kind":"message","messageId":"m1",
-			"parts":[{"kind":"text","text":"ping"}]}}}`)
-	if task.Status.State != stateCompleted || textOf(task.Artifacts[0].Parts) != "pong" {
-		t.Fatalf("skill task = %+v, want completed pong", task)
+	if card.ProtocolVersion == "" || len(card.Skills) == 0 {
+		t.Errorf("card missing protocolVersion or skills: %+v", card)
 	}
 }

@@ -676,11 +644,3 @@ func rpcTask(t *testing.T, url, body string) Task {
 	}
 	return resp.Result
 }
-
-func skillIDs(skills []Skill) []string {
-	ids := make([]string, 0, len(skills))
-	for _, skill := range skills {
-		ids = append(ids, skill.ID)
-	}
-	return ids
-}
@@ -1,129 +0,0 @@
-package mcp
-
-import (
-	"encoding/json"
-	"net/http"
-	"strings"
-)
-
-// HandlerOption configures NewHandler.
-type HandlerOption func(*handlerOptions)
-
-type handlerOptions struct {
-	serverName, serverVersion, protocolVersion string
-}
-
-// WithServerInfo sets the name/version advertised in the initialize response.
-func WithServerInfo(name, version string) HandlerOption {
-	return func(o *handlerOptions) { o.serverName, o.serverVersion = name, version }
-}
-
-// WithProtocolVersion sets the MCP protocol version advertised in initialize.
-func WithProtocolVersion(v string) HandlerOption {
-	return func(o *handlerOptions) { o.protocolVersion = v }
-}
-
-// NewHandler returns an http.Handler serving the MCP protocol over HTTP as
-// JSON-RPC 2.0 (initialize, ping, notifications/*, tools/list, tools/call),
-// backed by the resolver. Mount it on your own server (e.g. POST /mcp): the
-// gateway provides the protocol; you keep your routes, middleware and any
-// human-facing docs page.
-func NewHandler(r Resolver, opts ...HandlerOption) http.Handler {
-	o := handlerOptions{serverName: "go-micro-mcp", serverVersion: "1.0.0", protocolVersion: "2024-11-05"}
-	for _, fn := range opts {
-		fn(&o)
-	}
-	return http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
-		if req.Method != http.MethodPost {
-			http.Error(w, "method not allowed", http.StatusMethodNotAllowed)
-			return
-		}
-		var rpc struct {
-			JSONRPC string          `json:"jsonrpc"`
-			ID      json.RawMessage `json:"id"`
-			Method  string          `json:"method"`
-			Params  json.RawMessage `json:"params"`
-		}
-		if err := json.NewDecoder(req.Body).Decode(&rpc); err != nil {
-			writeRPCError(w, nil, ParseError, "Parse error", err.Error())
-			return
-		}
-
-		// Notifications (and any id-less request) expect no response body.
-		if strings.HasPrefix(rpc.Method, "notifications/") || len(rpc.ID) == 0 {
-			w.WriteHeader(http.StatusNoContent)
-			return
-		}
-
-		ctx := req.Context()
-		switch rpc.Method {
-		case "initialize":
-			writeRPCResult(w, rpc.ID, map[string]interface{}{
-				"protocolVersion": o.protocolVersion,
-				"capabilities":    map[string]interface{}{"tools": map[string]interface{}{}},
-				"serverInfo":      map[string]interface{}{"name": o.serverName, "version": o.serverVersion},
-			})
-		case "ping":
-			writeRPCResult(w, rpc.ID, map[string]interface{}{})
-		case "tools/list":
-			tools, err := r.List(ctx)
-			if err != nil {
-				writeRPCError(w, rpc.ID, InternalError, "Failed to list tools", err.Error())
-				return
-			}
-			list := make([]map[string]interface{}, 0, len(tools))
-			for _, t := range tools {
-				list = append(list, map[string]interface{}{
-					"name": t.Name, "description": t.Description, "inputSchema": t.InputSchema,
-				})
-			}
-			writeRPCResult(w, rpc.ID, map[string]interface{}{"tools": list})
-		case "tools/call":
-			var p struct {
-				Name      string                 `json:"name"`
-				Arguments map[string]interface{} `json:"arguments"`
-			}
-			if err := json.Unmarshal(rpc.Params, &p); err != nil {
-				writeRPCError(w, rpc.ID, InvalidParams, "Invalid params", err.Error())
-				return
-			}
-			res, err := r.Call(ctx, p.Name, p.Arguments)
-			if err != nil {
-				// Protocol/pre-check failure -> JSON-RPC error. An *RPCError
-				// carries a specific code; anything else is InternalError.
-				if rpcErr, ok := err.(*RPCError); ok {
-					writeRPCError(w, rpc.ID, rpcErr.Code, rpcErr.Message, rpcErr.Data)
-				} else {
-					writeRPCError(w, rpc.ID, InternalError, "Tool call failed", err.Error())
-				}
-				return
-			}
-			result := map[string]interface{}{
-				"content": []map[string]interface{}{{"type": "text", "text": res.Text}},
-			}
-			if res.IsError {
-				result["isError"] = true
-			}
-			writeRPCResult(w, rpc.ID, result)
-		default:
-			writeRPCError(w, rpc.ID, MethodNotFound, "Method not found", rpc.Method)
-		}
-	})
-}
-
-func writeRPCResult(w http.ResponseWriter, id json.RawMessage, result interface{}) {
-	w.Header().Set("Content-Type", "application/json")
-	_ = json.NewEncoder(w).Encode(map[string]interface{}{"jsonrpc": "2.0", "id": rawOrNull(id), "result": result})
-}
-
-func writeRPCError(w http.ResponseWriter, id json.RawMessage, code int, msg string, data interface{}) {
-	w.Header().Set("Content-Type", "application/json")
-	_ = json.NewEncoder(w).Encode(map[string]interface{}{"jsonrpc": "2.0", "id": rawOrNull(id), "error": map[string]interface{}{"code": code, "message": msg, "data": data}})
-}
-
-func rawOrNull(id json.RawMessage) interface{} {
-	if len(id) == 0 {
-		return nil
-	}
-	return id
-}
@@ -1,131 +0,0 @@
-package mcp
-
-import (
-	"context"
-	"sync"
-
-	"go-micro.dev/v6/ai"
-	"go-micro.dev/v6/client"
-	"go-micro.dev/v6/registry"
-)
-
-// CallResult is the outcome of a successful tool dispatch. A tool that ran but
-// produced an error sets IsError — per the MCP spec this is returned as a
-// tools/call result with isError:true, not a JSON-RPC protocol error.
-type CallResult struct {
-	Text    string
-	IsError bool
-}
-
-// Error lets the package's RPCError (see stdio.go) be returned by a resolver
-// to signal a protocol/pre-check failure with a specific JSON-RPC code; the
-// handler maps it straight to the JSON-RPC error.
-func (e *RPCError) Error() string { return e.Message }
-
-// ToolFunc executes a manually-registered tool. Return a *CallResult for tool
-// outcomes (set IsError for tool-level failures); return a non-nil error — an
-// *RPCError for a specific code — for protocol/pre-check failures.
-type ToolFunc func(ctx context.Context, args map[string]any) (*CallResult, error)
-
-// Resolver supplies the gateway's tools and executes calls. Swapping the
-// resolver changes where tools come from without touching the MCP protocol or
-// transport:
-//
-//   - NewManualResolver:   tools you register explicitly (full product control,
-//     including tools that are not go-micro services, executed via your own
-//     logic — auth, metering, …).
-//   - NewRegistryResolver: tools auto-discovered from registered services.
-//
-// The built-in store/broker tools are intentionally NOT exposed by any
-// resolver — they remain a development convenience on the legacy Serve() path.
-type Resolver interface {
-	// List returns the current tool catalog.
-	List(ctx context.Context) ([]Tool, error)
-	// Call executes a tool by name with JSON arguments.
-	Call(ctx context.Context, name string, args map[string]any) (*CallResult, error)
-}
-
-// ManualResolver exposes an explicitly-registered set of tools.
-type ManualResolver struct {
-	mu    sync.RWMutex
-	order []Tool
-	funcs map[string]ToolFunc
-}
-
-// NewManualResolver returns an empty manual resolver.
-func NewManualResolver() *ManualResolver {
-	return &ManualResolver{funcs: map[string]ToolFunc{}}
-}
-
-// Add registers (or replaces) a tool and its handler. Returns the resolver for
-// chaining.
-func (m *ManualResolver) Add(t Tool, fn ToolFunc) *ManualResolver {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-	if _, ok := m.funcs[t.Name]; ok {
-		for i := range m.order {
-			if m.order[i].Name == t.Name {
-				m.order[i] = t
-			}
-		}
-	} else {
-		m.order = append(m.order, t)
-	}
-	m.funcs[t.Name] = fn
-	return m
-}
-
-// List returns the registered tools.
-func (m *ManualResolver) List(_ context.Context) ([]Tool, error) {
-	m.mu.RLock()
-	defer m.mu.RUnlock()
-	out := make([]Tool, len(m.order))
-	copy(out, m.order)
-	return out, nil
-}
-
-// Call runs the handler registered for name.
-func (m *ManualResolver) Call(ctx context.Context, name string, args map[string]any) (*CallResult, error) {
-	m.mu.RLock()
-	fn, ok := m.funcs[name]
-	m.mu.RUnlock()
-	if !ok {
-		return nil, &RPCError{Code: InvalidParams, Message: "Tool not found: " + name, Data: name}
-	}
-	return fn(ctx, args)
-}
-
-// RegistryResolver auto-discovers tools from registered go-micro services and
-// executes them over RPC. It exposes only services — never the internal
-// store/broker tools.
-type RegistryResolver struct {
-	tools *ai.Tools
-}
-
-// NewRegistryResolver discovers services from reg and calls them with cl.
-func NewRegistryResolver(reg registry.Registry, cl client.Client) *RegistryResolver {
-	return &RegistryResolver{tools: ai.NewTools(reg, ai.ToolClient(cl))}
-}
-
-// List discovers the current service tools.
-func (r *RegistryResolver) List(_ context.Context) ([]Tool, error) {
-	discovered, err := r.tools.Discover()
-	if err != nil {
-		return nil, err
-	}
-	out := make([]Tool, 0, len(discovered))
-	for _, t := range discovered {
-		out = append(out, Tool{
-			Name:        t.Name,
-			Description: t.Description,
-			InputSchema: map[string]interface{}{"type": "object", "properties": t.Properties},
-		})
-	}
-	return out, nil
-}
-
-// Call executes a discovered service tool.
-func (r *RegistryResolver) Call(ctx context.Context, name string, args map[string]any) (*CallResult, error) {
-	res := r.tools.Handler()(ctx, ai.ToolCall{ID: "1", Name: name, Input: args})
-	return &CallResult{Text: res.Content}, nil
-}
@@ -1,64 +0,0 @@
-package mcp
-
-import (
-	"context"
-	"encoding/json"
-	"net/http"
-	"net/http/httptest"
-	"strings"
-	"testing"
-)
-
-func TestManualResolverHandler(t *testing.T) {
-	res := NewManualResolver().
-		Add(Tool{Name: "echo", Description: "echoes text"},
-			func(_ context.Context, args map[string]interface{}) (*CallResult, error) {
-				s, _ := args["text"].(string)
-				return &CallResult{Text: "you said: " + s}, nil
-			}).
-		Add(Tool{Name: "boom", Description: "errors"},
-			func(_ context.Context, _ map[string]interface{}) (*CallResult, error) {
-				return &CallResult{Text: "kaboom", IsError: true}, nil
-			}).
-		Add(Tool{Name: "blocked", Description: "coded error"},
-			func(_ context.Context, _ map[string]interface{}) (*CallResult, error) {
-				return nil, &RPCError{Code: -32000, Message: "insufficient credits"}
-			})
-
-	ts := httptest.NewServer(NewHandler(res))
-	defer ts.Close()
-	rpc := func(body string) (int, map[string]interface{}) {
-		resp, err := http.Post(ts.URL, "application/json", strings.NewReader(body))
-		if err != nil {
-			t.Fatalf("post rpc: %v", err)
-		}
-		defer resp.Body.Close()
-		var out map[string]interface{}
-		json.NewDecoder(resp.Body).Decode(&out)
-		return resp.StatusCode, out
-	}
-
-	if _, out := rpc(`{"jsonrpc":"2.0","id":1,"method":"tools/list"}`); len(out["result"].(map[string]interface{})["tools"].([]interface{})) != 3 {
-		t.Fatalf("tools/list: %v", out)
-	}
-	// tool result
-	_, out := rpc(`{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"echo","arguments":{"text":"hi"}}}`)
-	if out["result"].(map[string]interface{})["content"].([]interface{})[0].(map[string]interface{})["text"] != "you said: hi" {
-		t.Fatalf("echo: %v", out)
-	}
-	// tool-level error -> isError result, NOT protocol error
-	_, out = rpc(`{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"boom","arguments":{}}}`)
-	if out["error"] != nil || out["result"].(map[string]interface{})["isError"] != true {
-		t.Fatalf("boom should be isError result: %v", out)
-	}
-	// coded protocol error -> JSON-RPC error with the code
-	_, out = rpc(`{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"blocked","arguments":{}}}`)
-	if out["error"] == nil || int(out["error"].(map[string]interface{})["code"].(float64)) != -32000 {
-		t.Fatalf("blocked should be -32000: %v", out)
-	}
-	// notification -> 204, no body
-	code, _ := rpc(`{"jsonrpc":"2.0","method":"notifications/initialized"}`)
-	if code != http.StatusNoContent {
-		t.Fatalf("notification status = %d, want 204", code)
-	}
-}
@@ -122,19 +122,6 @@ database "agent", table "{name}":
  history    — conversation history
 ```

-## Durable Ask / StreamAsk runs
-
-Agents can opt into the same checkpoint backend used by flows with
-`micro.AgentWithCheckpoint(...)`. When enabled, each `Ask` or `StreamAsk` run is
-persisted with its input, terminal status, response, and tool-call records. If
-the process or transport drops after a tool has completed but before the model
-returns a final answer, restart the agent with the same checkpoint store and
-call `micro.AgentResume(ctx, ag, runID)` or
-`micro.AgentResumeStreamAsk(ctx, ag, runID)`.
-Completed tool calls are served from the checkpoint instead of being executed
-again, while guardrails such as `MaxSteps`, loop detection, approval pauses, and
-`request_input` pauses continue to apply to the resumed run.
-
 ## Built-in Capabilities

 Beyond its scoped service tools, every agent gets two built-in tools. They are not service endpoints — they are capabilities the agent has over itself and over other agents. They are plain tools wired into the agent's tool handler; there is no separate harness, loop engine, or graph. The LLM calls them exactly like any other tool.
@@ -19,10 +19,11 @@ redirect the loop; direction always wins.
 items the loop can auto-merge): brand/positioning copy, breaking public-API
 changes, architectural rewrites. Those go to the human.

-## Work queue (ranked)
+## Developer experience (ranked)

-1. **Harden agent loop failure handling** ([#3391](https://github.com/micro/go-micro/issues/3391)) — now that scheduled cross-provider agent conformance shipped, close the next Now-phase reliability gap: context deadlines, cancellation, provider timeouts/rate limits, retry/backoff boundaries, and safe mid-run failure semantics across model calls, tool execution, plan/delegate, and guardrails. This makes the service-as-tool harness operable under real provider conditions before layering more agentic depth on top.
-2. **CI-verify 0-to-1 and 0-to-hero developer flows** ([#3392](https://github.com/micro/go-micro/issues/3392)) — preserve the getting-started contract as an executable harness covering scaffold → run → call plus the multi-service/agent path through chat, inspection, and deploy dry-run where practical. This keeps the CLI-first inner loop aligned with the README, website, and roadmap while the harness evolves.
+1. **Schedule cross-provider agent conformance** ([#3295](https://github.com/micro/go-micro/issues/3295)) — after the 0→hero reference and provider-focused unit coverage shipped, the highest-value remaining Now-roadmap hardening gap is a scheduled, key-gated provider matrix that proves the same agent/tool workflow keeps working across supported models without blocking contributors who lack secrets.
+2. **Harden agent failure and cancellation semantics** ([#3296](https://github.com/micro/go-micro/issues/3296)) — the harness is increasingly durable, streaming, observable, and human-in-the-loop; the next operability seam is making timeouts, cancellation, rate-limit errors, and retry/backoff behavior predictable across agent, AI provider, service-tool, and flow boundaries.
+3. **Expose run inspection in the CLI inner loop** ([#3297](https://github.com/micro/go-micro/issues/3297)) — the canon promises scaffold → run → chat → inspect → deploy, and recent work has strengthened scaffold/run/chat/deploy; inspection remains the most visible DX gap for turning agent/flow activity into actionable breadcrumbs during local development.

 _Seeded by Claude Code from the roadmap + open issues; thereafter maintained by the
 architecture-review pass._
@@ -7,10 +7,9 @@ suite is safe for local development, forks, and scheduled CI.

 ## What it exercises

-`go run ./internal/harness/provider-conformance` fans out over the provider-facing
-agent test and the harnesses in `internal/harness`:
+`go run ./internal/harness/provider-conformance` fans out over the harnesses in
+`internal/harness`:

- `agent` — provider tool-call conformance through `agent.Ask`, including run metadata propagation.
 - `universe` — service discovery plus agent tool calls over the real runtime.
 - `agent-flow` — a workflow event that drives an agent to call services.
 - `plan-delegate` — plan persistence plus agent-to-agent delegation and service
@@ -60,14 +59,13 @@ go run ./internal/harness/provider-conformance \
 ## Scheduled CI behavior

 The `Harness (E2E)` workflow runs on pushes and pull requests with deterministic
-mock LLMs, including `provider-conformance -providers mock`. On the daily schedule and manual dispatch it also runs the live
+mock LLMs. On the daily schedule and manual dispatch it also runs the live
 provider conformance job. That job:

-1. runs the same `agent`, `universe`, `agent-flow`, and `plan-delegate` harness list,
-2. reads the provider keys from repository secrets,
-3. skips providers whose secrets are absent,
-4. fails when any configured provider fails a harness, and
-5. uploads JSON and Markdown coverage artifacts for the run.
+1. reads the provider keys from repository secrets,
+2. skips providers whose secrets are absent,
+3. fails when any configured provider fails a harness, and
+4. uploads JSON and Markdown coverage artifacts for the run.

 The job also appends the Markdown summary and capability matrix to the GitHub
 Actions step summary, making configured, skipped, and failed provider coverage
@@ -46,7 +46,7 @@ var providerEnv = map[string]string{

 func main() {
 	providersFlag := flag.String("providers", "anthropic,openai,gemini,groq,mistral,together,atlascloud", "comma-separated providers to check; use mock for deterministic local checks")
-	harnessesFlag := flag.String("harnesses", "agent,universe,agent-flow,plan-delegate", "comma-separated harness names under internal/harness; agent runs the provider tool-call conformance test")
+	harnessesFlag := flag.String("harnesses", "universe,agent-flow,plan-delegate", "comma-separated harness names under internal/harness")
 	timeoutFlag := flag.Duration("timeout", 10*time.Minute, "timeout per provider/harness run")
 	requireConfiguredFlag := flag.Bool("require-configured", false, "fail when a selected live provider is missing an API key")
 	capabilitiesFlag := flag.Bool("capabilities", true, "print the registered provider capability matrix before running conformance")
@@ -244,9 +244,6 @@ func validateSelection(providers, harnesses []string) error {
 	}

 	for _, harness := range harnesses {
-		if harness == "agent" {
-			continue
-		}
 		if strings.Contains(harness, string(os.PathSeparator)) || harness == "." || harness == ".." {
 			return fmt.Errorf("invalid harness name %q", harness)
 		}
@@ -326,10 +323,6 @@ func localRPCEnv(env []string) []string {
 }

 func runHarness(provider, harness string, timeout time.Duration) error {
-	if harness == "agent" {
-		return runAgentConformance(provider, timeout)
-	}
-
 	ctx, cancel := context.WithTimeout(context.Background(), timeout)
 	defer cancel()

@@ -365,28 +358,3 @@ func runHarness(provider, harness string, timeout time.Duration) error {
 	}
 	return nil
 }
-
-func runAgentConformance(provider string, timeout time.Duration) error {
-	ctx, cancel := context.WithTimeout(context.Background(), timeout)
-	defer cancel()
-
-	testProvider := provider
-	if provider == "mock" {
-		testProvider = "fake"
-	}
-	cmd := exec.CommandContext(ctx, "go", "test", "./agent", "-run", "TestAgentProviderConformanceMatrix", "-count=1", "-v")
-	cmd.Dir = repoRoot()
-	cmd.Stdout = os.Stdout
-	cmd.Stderr = os.Stderr
-	cmd.Env = localRPCEnv(append(os.Environ(),
-		"GO_MICRO_AGENT_CONFORMANCE_LIVE=1",
-		"GO_MICRO_AGENT_CONFORMANCE_PROVIDERS="+testProvider,
-	))
-	if err := cmd.Run(); err != nil {
-		if ctx.Err() == context.DeadlineExceeded {
-			return fmt.Errorf("timed out after %s", timeout)
-		}
-		return err
-	}
-	return nil
-}
@@ -4,34 +4,13 @@ This directory owns the no-secret reference scenario for the Go Micro
 services → agents → workflows lifecycle. It is intentionally small and
 scripted so CI can run it on every push without external services or model keys.

-`run.sh` verifies four boundaries together:
+`run.sh` verifies three boundaries together:

 1. **Run** — `micro run` remains available as the local development entry point.
 2. **Chat** — `micro chat` remains available as the interactive agent entry point.
-3. **Inspect** — `micro inspect agent <name>` and `micro inspect flow <name>`
-   remain available as the local run-history inspection step, with `micro flow
-   runs` preserving durable workflow history inspection.
-4. **Deploy** — `micro deploy --dry-run <target>` remains available as the
-   deployment-boundary checkpoint. The dry run resolves configured deploy targets
-   and services and prints the remote build/copy/systemd/health plan without
-   building binaries, opening SSH connections, running `rsync`, or touching
-   remote infrastructure.
+3. **Inspect** — `micro flow runs` remains available for durable workflow run
+   history inspection.

 After the CLI boundary smoke checks, the script runs the deterministic harnesses
 that boot real services, agents, workflows, store-backed run history, and A2A
 with only the LLM mocked.
-
-## Local and CI entry points
-
-The default GitHub harness workflow runs this script on every push and pull
-request after the 0→1 scaffold contract. Developers can run the same no-secret
-contract locally with:
-
-```sh
-make harness
-```
-
-That target intentionally exercises the documented getting-started path before
-the 0→hero scenario, so the public scaffold → run/chat → inspect → deploy lifecycle stays
-executable outside CI as well. Live provider checks remain separate and gated by
-configured API keys (`make provider-conformance` or the scheduled/manual CI job).
@@ -7,7 +7,6 @@ cd "$ROOT"
 # Keep the developer inner-loop boundaries executable and discoverable in CI
 # without secrets or long-running daemons.
 go test ./cmd/micro -run TestZeroToHeroCLIBoundaries -count=1
-go test ./cmd/micro/cli/deploy -run TestDeployDryRun -count=1

 # Deterministic no-secret reference scenarios. These use the real Go Micro
 # runtime and mock only the LLM provider.
@@ -69,18 +69,10 @@ A card looks like:
  "capabilities": { "streaming": true, "pushNotifications": true },
  "defaultInputModes": ["text/plain"],
  "defaultOutputModes": ["text/plain"],
-  "skills": [
-    { "id": "task", "name": "Task", "tags": ["task"] },
-    { "id": "project", "name": "Project", "tags": ["project"] }
-  ]
+  "skills": [{ "id": "chat", "name": "Chat", "tags": ["task", "project"] }]
 }
 ```

-Each managed service is advertised as its own typed skill. Clients can call the
-whole agent at `/agents/task-mgr`, or address one skill directly at
-`/agents/task-mgr/skills/task`; the skill endpoint serves a focused card and
-routes the request to the same agent with that skill selected.
-
 ## Calling an agent

 A2A uses JSON-RPC 2.0 over HTTP. Send a message with `message/send`; the gateway runs the agent and returns a completed `Task`:
@@ -36,46 +36,13 @@ your stack — the harness *is* the stack.
 | Discovery & RPC | Registry + client; agents and services find and call each other | Shipped |
 | Interop | MCP (tools), A2A (agents), x402 (paid tools) | Shipped |
 | Resilience | Per-call timeout with context propagation; opt-in retry/backoff (`ModelRetry`) across the loop | Shipped |
-| Durable runs | Checkpoint and resume an agent run with the same checkpoint backend flows use | Shipped |
+| Durable runs | Checkpoint and resume an agent run (flows already do) | In progress |
 | Observability | `RunInfo` → OpenTelemetry spans for runs, model calls, tools, delegation, and failures; persisted run history | Shipped |
 | Streaming | `ai.Stream` through chat, agent, and A2A | In progress |

 The "in progress" rows are exactly the roadmap's [Now and Next](/docs/roadmap.html),
 and the work is happening in the open.

-## Durable agent runs
-
-Agents can persist their execution history to the same `Checkpoint` backend as
-flows. A checkpointed `Ask` records the run id, original prompt, model result,
-and completed tool calls. If the process restarts after a tool succeeds but
-before the model finishes, `AgentResume` continues the same run and returns the
-recorded tool result instead of re-running the side effect. If a run already
-completed, resume returns the persisted response without calling the model.
-
-```go
-agent := micro.NewAgent("conductor",
-    micro.AgentProvider("anthropic"),
-    micro.AgentWithCheckpoint(checkpoint),
-)
-
-resp, err := agent.Ask(ctx, "charge order 42 and send a receipt")
-if err != nil {
-    // On startup, or after a transient failure, discover unfinished work:
-    pending, _ := micro.AgentPending(ctx, agent)
-    for _, run := range pending {
-        _, _ = micro.AgentResume(ctx, agent, run.ID)
-    }
-}
-_ = resp
-```
-
-For human-in-the-loop runs that pause through the built-in `request_input` tool,
-resume with the operator's response:
-
-```go
-_, err := micro.AgentResumeInput(ctx, agent, runID, "Deploy to us-east-1")
-```
-
 ## Observing agent runs

 Pass an OpenTelemetry tracer provider when you construct an agent to turn the
@@ -83,44 +83,6 @@ a := micro.NewAgent("conductor",
 a.Ask(ctx, "Plan the launch, create the tasks, and have comms notify the owner.")
 ```

-### Long-running memory
-
-Agents use store-backed conversation memory by default, scoped under the agent's
-name. That makes short restarts boring: the next `Ask` reloads the retained
-history from the same store backend you already use for services and flows.
-Long-running agents can also keep model context bounded without losing useful
-prior context:
-
-```go
-a := micro.NewAgent("conductor",
-    micro.AgentServices("task"),
-    micro.AgentProvider("anthropic"),
-    micro.AgentCompactMemory(40, 12),      // max active messages, recent messages kept verbatim
-    micro.AgentMemoryRecallLimit(5),       // archived turns recalled per Ask
-)
-```
-
-`AgentCompactMemory(maxMessages, keepRecent)` switches the default memory to a
-deterministic compactor. Once active history grows past `maxMessages`, older
-turns move into the durable archive, a provider-neutral summary is injected into
-active context, and the newest `keepRecent` messages stay verbatim. On future
-asks, archived turns whose text matches the current request are recalled ahead of
-the active context. The built-in retrieval is intentionally simple and
-credential-free for CI; teams that need embeddings or a vector database can still
-provide their own `AgentMemory` implementation.
-
-This is harness memory, not prompt-layer orchestration: services remain the
-capabilities, agents remain the dynamic decision makers, and flows remain the
-durable predefined paths. Compaction only keeps a scheduled or looping agent from
-turning every past turn into model context while still letting it remember facts
-that matter to the current service → agent → workflow run.
-
-Checkpointed agent runs and compacted memory share the same store-backed shape.
-If a provider call fails after the prompt has been recorded, `agent.Resume` uses
-the checkpointed run id and does not append that same user turn a second time;
-completed tool results and recalled archived memory remain available for the
-retry.
-
 ## The patterns — most are already here

 Anthropic lists five workflow patterns. Go Micro implements the two richest ones natively, as services and tools, and the rest are ordinary compositions:
@@ -39,12 +39,12 @@ The built-in providers currently register these capability interfaces:
 | Provider | Chat/text (`ai.Model`) | Image (`ai.ImageModel`) | Video (`ai.VideoModel`) | Streaming (`ai.Stream`) |
 | --- | --- | --- | --- | --- |
 | `anthropic` | Yes | No | No | No |
-| `atlascloud` | Yes | Yes | Yes | Yes |
+| `atlascloud` | Yes | Yes | Yes | No |
 | `gemini` | Yes | No | No | No |
-| `groq` | Yes | No | No | Yes |
-| `mistral` | Yes | No | No | Yes |
+| `groq` | Yes | No | No | No |
+| `mistral` | Yes | No | No | No |
 | `openai` | Yes | Yes | No | Yes |
-| `together` | Yes | No | No | Yes |
+| `together` | Yes | No | No | No |

 ## Step 1: Implement the `ai.Model` Interface

@@ -36,13 +36,14 @@ The priority is that what exists works everywhere, under real conditions.

 ## Next — agentic depth

+- **Durable agent loop.** Flows resume; the agent's own loop does not yet. Reuse `Checkpoint` so a long-running agent survives a restart and continues.
 - **Streaming.** Broaden provider-backed `ai.Stream` coverage and keep chat plus A2A `message/stream` working end to end for real chat and long-task UX.
 - **Agent observability.** Wire the new `RunInfo` into OpenTelemetry spans so a run — steps, tool calls, delegation — is traceable. This is also what anyone running it in production will need.

 ## Later

 - **Memory management** — summarization and retrieval (RAG) beyond a fixed buffer.
- **Human-in-the-loop** — broaden pause/resume UX around `input-required` runs and approvals.
+- **Human-in-the-loop** — pause and resume mid-run (`input-required`), beyond the binary `ApproveTool` gate.
 - **A2A** — richer live-stream reconnection (`tasks/resubscribe`) and `input-required` handoffs.

 ## Developer experience (ongoing)
@@ -23,21 +23,12 @@ type Service = service.Service
 // Agent is the interface for an AI agent that manages services.
 type Agent = agent.Agent

-// AgentResponse is what an agent returns from Ask or a resumed run.
-type AgentResponse = agent.Response
-
-// AgentStream is a stream of tool execution events followed by final-answer chunks.
-type AgentStream = agent.AgentStream
-
 // AgentOption configures an Agent.
 type AgentOption = agent.Option

 // Flow is an event-driven LLM orchestration unit.
 type Flow = flow.Flow

-// FlowRun is a checkpointed flow or agent run record.
-type FlowRun = flow.Run
-
 // FlowOption configures a Flow.
 type FlowOption = flow.Option

@@ -173,32 +164,6 @@ func AgentWrapTool(w ...ai.ToolWrapper) AgentOption {
 // tool calls, delegation, and failures.
 func AgentTraceProvider(tp trace.TracerProvider) AgentOption { return agent.TraceProvider(tp) }

-// AgentWithCheckpoint sets the durability backend for agent Ask runs.
-// It uses the same Checkpoint interface as flows so services, agents,
-// and workflows can share one execution history backend.
-func AgentWithCheckpoint(c Checkpoint) AgentOption { return agent.WithCheckpoint(c) }
-
-// AgentPending returns checkpointed agent runs that have not completed.
-// Use it at process startup to discover agent work that should be resumed.
-func AgentPending(ctx context.Context, a Agent) ([]FlowRun, error) { return agent.Pending(ctx, a) }
-
-// AgentResume resumes a checkpointed agent run by id. Completed runs return
-// the persisted response without calling the model or replaying tool calls.
-func AgentResume(ctx context.Context, a Agent, runID string) (*AgentResponse, error) {
-	return agent.Resume(ctx, a, runID)
-}
-
-// AgentResumeInput resumes a checkpointed agent run waiting for human input.
-func AgentResumeInput(ctx context.Context, a Agent, runID, input string) (*AgentResponse, error) {
-	return agent.ResumeInput(ctx, a, runID, input)
-}
-
-// AgentResumeStreamAsk resumes a checkpointed agent run by id and streams the
-// resulting tool events and final answer.
-func AgentResumeStreamAsk(ctx context.Context, a Agent, runID string) (AgentStream, error) {
-	return agent.ResumeStreamAsk(ctx, a, runID)
-}
-
 // NewFlow creates an event-driven LLM orchestration unit.
 //
 //	f := micro.NewFlow("onboard-user",
Author	SHA1	Message	Date
Asim Aslam	d40d52b891	ai/openai: surface token usage on streams Harness (E2E) / Harnesses (mock LLM) (push) Waiting to run Details Harness (E2E) / Provider harnesses (live LLM conformance) (push) Waiting to run Details Lint / golangci-lint (push) Waiting to run Details Run Tests / Unit Tests (push) Waiting to run Details Run Tests / Etcd Integration Tests (push) Waiting to run Details Request stream_options.include_usage and return the final usage chunk as a Response with Usage set, so streaming callers can record usage.	2026-06-28 23:13:22 +01:00
Asim Aslam	eefc30c569	ai/atlascloud: surface token usage on streams Request stream_options.include_usage and return the final usage chunk as a Response with Usage set, so streaming callers can record usage.	2026-06-28 23:13:21 +01:00