chore: import upstream snapshot with attribution

2026-07-13 12:33:42 +08:00
commit a06f331eb8
3186 changed files with 689843 additions and 0 deletions
@@ -0,0 +1,129 @@
+package main
+
+import (
+	"github.com/spf13/cobra"
+)
+
+// feedbackDaemonTool is the daemon-tool relay seam. It is indirected through a
+// package var so tests can stub the daemon call (asserting the lowered
+// tool + args) without a running daemon.
+var feedbackDaemonTool = requireDaemonTool
+
+// Shared persistent flags for the feedback group.
+var (
+	feedbackIndex  string
+	feedbackFormat string
+)
+
+var feedbackCmd = &cobra.Command{
+	Use:   "feedback",
+	Short: "Agent-learning feedback — score & query smart_context suggestions (feedback)",
+	Long: `The feedback verb group records and queries the agent-learning feedback
+loop over the daemon:
+
+  record — mark which smart_context / prefetch_context suggestions were useful,
+           not needed, or missing (improves future context quality)
+  query  — aggregated stats: most useful, most missed, accuracy
+
+Every subcommand is a thin shell over the MCP feedback tool on the daemon that
+owns the repo. Requires a running daemon that tracks the repo.`,
+	SilenceUsage: true,
+}
+
+// runFeedbackTool calls the feedback tool and renders the result. With --format
+// json (or as the fallback for any shape) it pretty-prints the JSON.
+func runFeedbackTool(cmd *cobra.Command, args map[string]any) error {
+	if feedbackFormat != "" {
+		args["format"] = feedbackFormat
+	}
+	raw, err := feedbackDaemonTool(feedbackIndex, "feedback", args)
+	if err != nil {
+		return err
+	}
+	return emitDaemonJSON(cmd, raw)
+}
+
+// --- feedback record --------------------------------------------------------
+
+var (
+	feedbackRecordTask       string
+	feedbackRecordUseful     string
+	feedbackRecordNotNeeded  string
+	feedbackRecordMissing    string
+	feedbackRecordToolSource string
+)
+
+var feedbackRecordCmd = &cobra.Command{
+	Use:   "record",
+	Short: "Report which context suggestions were useful / not needed / missing",
+	Long: `Records feedback on a prior smart_context / prefetch_context call: the
+comma-separated symbol IDs that were useful, returned-but-not-needed, or that
+should have been included. This trains future context quality.
+
+  gortex feedback record --task "fix the auth bug" --useful pkg/a.go::Foo,pkg/b.go::Bar`,
+	RunE: func(cmd *cobra.Command, _ []string) error {
+		args := map[string]any{"action": "record"}
+		if cmd.Flags().Changed("task") {
+			args["task"] = feedbackRecordTask
+		}
+		if cmd.Flags().Changed("useful") {
+			args["useful"] = feedbackRecordUseful
+		}
+		if cmd.Flags().Changed("not-needed") {
+			args["not_needed"] = feedbackRecordNotNeeded
+		}
+		if cmd.Flags().Changed("missing") {
+			args["missing"] = feedbackRecordMissing
+		}
+		if cmd.Flags().Changed("tool-source") {
+			args["tool_source"] = feedbackRecordToolSource
+		}
+		return runFeedbackTool(cmd, args)
+	},
+}
+
+// --- feedback query ---------------------------------------------------------
+
+var (
+	feedbackQueryToolSource string
+	feedbackQueryTopN       int
+	feedbackQueryCompact    bool
+)
+
+var feedbackQueryCmd = &cobra.Command{
+	Use:   "query",
+	Short: "Aggregated feedback stats — most useful, most missed, accuracy",
+	RunE: func(cmd *cobra.Command, _ []string) error {
+		args := map[string]any{"action": "query"}
+		if cmd.Flags().Changed("tool-source") {
+			args["tool_source"] = feedbackQueryToolSource
+		}
+		if cmd.Flags().Changed("top-n") {
+			args["top_n"] = feedbackQueryTopN
+		}
+		if cmd.Flags().Changed("compact") {
+			args["compact"] = feedbackQueryCompact
+		}
+		return runFeedbackTool(cmd, args)
+	},
+}
+
+func init() {
+	feedbackCmd.PersistentFlags().StringVar(&feedbackIndex, "index", ".", "repository path the daemon must track")
+	feedbackCmd.PersistentFlags().StringVar(&feedbackIndex, "repo", ".", "alias for --index")
+	feedbackCmd.PersistentFlags().StringVar(&feedbackFormat, "format", "json", "output / wire format: json|gcx|toon|text")
+
+	feedbackRecordCmd.Flags().StringVar(&feedbackRecordTask, "task", "", "the task description used in the original context call")
+	feedbackRecordCmd.Flags().StringVar(&feedbackRecordUseful, "useful", "", "comma-separated symbol IDs that were useful")
+	feedbackRecordCmd.Flags().StringVar(&feedbackRecordNotNeeded, "not-needed", "", "comma-separated symbol IDs returned but not needed (not_needed)")
+	feedbackRecordCmd.Flags().StringVar(&feedbackRecordMissing, "missing", "", "comma-separated symbol IDs that should have been included")
+	feedbackRecordCmd.Flags().StringVar(&feedbackRecordToolSource, "tool-source", "", "which tool produced the context: smart_context (default) or prefetch_context (tool_source)")
+
+	feedbackQueryCmd.Flags().StringVar(&feedbackQueryToolSource, "tool-source", "", "filter by source: smart_context, prefetch_context, or all (tool_source)")
+	feedbackQueryCmd.Flags().IntVar(&feedbackQueryTopN, "top-n", 0, "number of top symbols per category (default 10) (top_n)")
+	feedbackQueryCmd.Flags().BoolVar(&feedbackQueryCompact, "compact", false, "one-line-per-symbol text output")
+
+	feedbackCmd.AddCommand(feedbackRecordCmd)
+	feedbackCmd.AddCommand(feedbackQueryCmd)
+	rootCmd.AddCommand(feedbackCmd)
+}