diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
new file mode 100644
index 00000000..cd65d486
--- /dev/null
+++ b/.github/workflows/test.yml
@@ -0,0 +1,16 @@
+name: test
+
+on:
+  push:
+  pull_request:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: python -m pip install --upgrade pip pyyaml
+      - run: make test
diff --git a/Makefile b/Makefile
index 1c342811..3fc9e88b 100644
--- a/Makefile
+++ b/Makefile
@@ -1,17 +1,26 @@
 PYTHON ?= python3
 
-.PHONY: eval package-check package-failure-check test clean
+.PHONY: eval eval-suite package-check package-failure-check validate lint test clean
 
 eval:
 	$(PYTHON) scripts/trigger_eval.py --description-file evals/improved_description.txt --cases evals/trigger_cases.json --baseline-description-file evals/baseline_description.txt
 
+eval-suite:
+	$(PYTHON) scripts/run_eval_suite.py
+
 package-check:
 	$(PYTHON) scripts/cross_packager.py . --platform openai --platform claude --platform generic --expectations evals/packaging_expectations.json --output-dir dist --zip
 
 package-failure-check:
 	$(PYTHON) tests/verify_packager_failures.py
 
-test: eval package-check package-failure-check
+validate:
+	$(PYTHON) scripts/validate_skill.py .
+
+lint:
+	$(PYTHON) scripts/lint_skill.py .
+
+test: eval eval-suite package-check package-failure-check validate lint
 
 clean:
 	rm -rf dist tests/tmp
diff --git a/README.md b/README.md
index e0991217..edc8980f 100644
--- a/README.md
+++ b/README.md
@@ -103,8 +103,10 @@ Long-form material that should not bloat the main skill file. This includes desi
 Utility scripts that make the meta-skill operational:
 
 - `trigger_eval.py`: evaluates trigger descriptions with positive, negative, and near-neighbor prompts
+- `run_eval_suite.py`: runs train/dev/holdout trigger suites and fails if aggregate regressions appear
 - `context_sizer.py`: estimates context weight and warns when the initial load gets too large
 - `cross_packager.py`: builds client-specific export artifacts with explicit platform contracts and validation
+- `init_skill.py`, `lint_skill.py`, `validate_skill.py`, `diff_eval.py`: minimal authoring toolchain
 
 ### `evals/`
 
@@ -114,10 +116,15 @@ Reusable trigger and packaging checks, including baseline and improved descripti
 
 Three end-to-end examples showing raw workflow input, design summary, and final generated skill shape.
 
+### `.github/workflows/test.yml`
+
+Continuous integration entrypoint that runs the full local regression suite on push and pull request.
+
 ## Validation Notes
 
 - Trigger evaluation is stronger than the original overlap-only version, but it is still heuristic.
 - The sample trigger report now covers a larger positive, negative, and near-neighbor set rather than a tiny demo set.
+- Train/dev/holdout trigger suites now separate iterative tuning from final verification.
 - Packaging validation now uses explicit contracts and YAML parsing, but it is still a lightweight local validation layer rather than a full platform integration suite.
 - `evals/failure-cases.md` captures known weak spots that should remain part of regression checks.
 - `tests/verify_packager_failures.py` checks that invalid metadata, invalid YAML, and unsupported targets fail clearly.
@@ -192,6 +199,7 @@ This project is best for:
 - Evals: [evals/README.md](evals/README.md)
 - Packaging contracts: [references/packaging-contracts.md](references/packaging-contracts.md)
 - Failure fixtures: [tests/fixtures](tests/fixtures)
+- Evolution example: [examples/evolution-frontend-review/README.md](examples/evolution-frontend-review/README.md)
 
 ## License
 
diff --git a/evals/README.md b/evals/README.md
index 553f6f7b..7462c910 100644
--- a/evals/README.md
+++ b/evals/README.md
@@ -5,6 +5,7 @@ This directory makes trigger quality and packaging quality more reproducible.
 Contents:
 
 - `trigger_cases.json`: positive, negative, and near-neighbor prompts
+- `train/`, `dev/`, `holdout/`: split trigger suites for iterative tuning and final verification
 - `baseline_description.txt`: intentionally weaker trigger description
 - `improved_description.txt`: current stronger trigger description
 - `sample_trigger_report.json`: example comparison output using threshold `0.35`
@@ -16,6 +17,7 @@ Use:
 ```bash
 python3 scripts/trigger_eval.py --description-file evals/improved_description.txt --cases evals/trigger_cases.json --threshold 0.35
 python3 scripts/trigger_eval.py --description-file evals/improved_description.txt --cases evals/trigger_cases.json --baseline-description-file evals/baseline_description.txt --threshold 0.35
+python3 scripts/run_eval_suite.py
 python3 scripts/cross_packager.py . --platform openai --platform claude --expectations evals/packaging_expectations.json --zip
 python3 tests/verify_packager_failures.py
 ```
@@ -28,3 +30,4 @@ Regression scope now includes:
 - long-context positives
 - mixed-intent negatives
 - explicit "do not build a skill" negatives
+- holdout verification
diff --git a/evals/dev/trigger_cases.json b/evals/dev/trigger_cases.json
new file mode 100644
index 00000000..ecacefb4
--- /dev/null
+++ b/evals/dev/trigger_cases.json
@@ -0,0 +1,27 @@
+{
+  "recommended_threshold": 0.37,
+  "negative_patterns": [
+    "one-off",
+    "do not turn",
+    "without building",
+    "brainstorm",
+    "just explain",
+    "headline",
+    "summarize"
+  ],
+  "should_trigger": [
+    "Convert this operations checklist into a reusable skill.",
+    "Add trigger evals to this skill before sharing it with the team.",
+    "We have a messy release runbook, export process, and a prompt history; turn all of that into one reusable skill package with evals and packaging checks."
+  ],
+  "should_not_trigger": [
+    "Give me ideas for improving our process.",
+    "Review this note and tell me what it means in plain English.",
+    "Help me brainstorm several ways to improve packaging, but do not generate any skill files."
+  ],
+  "near_neighbor": [
+    "Review this process note and explain it, no packaging needed.",
+    "Turn this into a checklist and template, but stop short of making a full skill.",
+    "Package this explanation as a document, not as an agent skill."
+  ]
+}
diff --git a/evals/holdout/trigger_cases.json b/evals/holdout/trigger_cases.json
new file mode 100644
index 00000000..bf791bd5
--- /dev/null
+++ b/evals/holdout/trigger_cases.json
@@ -0,0 +1,23 @@
+{
+  "recommended_threshold": 0.37,
+  "negative_patterns": [
+    "one-off",
+    "do not turn",
+    "without building",
+    "brainstorm",
+    "just explain",
+    "headline",
+    "summarize"
+  ],
+  "should_trigger": [
+    "Build a reusable skill from this long internal process note, and make sure it has references, scripts, and a trigger description.",
+    "Take this existing skill draft, tighten the trigger boundary, add near-neighbor evals, and package it for distribution."
+  ],
+  "should_not_trigger": [
+    "I pasted a long process description below, but I only want a summary, not a reusable skill."
+  ],
+  "near_neighbor": [
+    "Help me shape an idea before we decide whether to build a skill.",
+    "Create an outline for a possible future skill, but do not build the skill yet."
+  ]
+}
diff --git a/evals/train/trigger_cases.json b/evals/train/trigger_cases.json
new file mode 100644
index 00000000..aa11f6cd
--- /dev/null
+++ b/evals/train/trigger_cases.json
@@ -0,0 +1,35 @@
+{
+  "recommended_threshold": 0.37,
+  "negative_patterns": [
+    "one-off",
+    "do not turn",
+    "without building",
+    "brainstorm",
+    "just explain",
+    "headline",
+    "summarize"
+  ],
+  "should_trigger": [
+    "Create a skill from this repeated workflow.",
+    "Improve this skill description and add evals.",
+    "Turn this runbook into a reusable agent skill.",
+    "Package this skill for team reuse.",
+    "Refactor this prompt into a proper skill package.",
+    "Create a meta-skill for packaging internal workflows."
+  ],
+  "should_not_trigger": [
+    "Explain what a workflow is.",
+    "Write a product headline for this landing page.",
+    "Summarize this random note.",
+    "Just explain what a skill is.",
+    "Draft a blog title for this article.",
+    "Translate this README into Japanese."
+  ],
+  "near_neighbor": [
+    "Create a one-off prompt for this task.",
+    "Help me brainstorm process ideas without building a skill.",
+    "Improve this README but do not turn it into a skill.",
+    "Make a checklist for this task, but not a reusable skill.",
+    "Write a custom answer for this request without creating a skill package."
+  ]
+}
diff --git a/examples/README.md b/examples/README.md
index ddcb77ec..97479134 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -13,3 +13,7 @@ Each example contains:
 - `raw-workflow.md`: what the user originally gives
 - `design-summary.md`: how the meta-skill scopes and structures the result
 - `generated-skill/`: the resulting skill package shape
+
+Additional evolution example:
+
+- `evolution-frontend-review/`: shows raw workflow, v0, v1, final, and eval delta
diff --git a/examples/evolution-frontend-review/README.md b/examples/evolution-frontend-review/README.md
new file mode 100644
index 00000000..394d5ee6
--- /dev/null
+++ b/examples/evolution-frontend-review/README.md
@@ -0,0 +1,9 @@
+# Evolution Example: Frontend Review
+
+This example shows a full improvement chain:
+
+- `raw-workflow.md`: initial request
+- `v0/`: initial skill draft
+- `v1/`: refined skill draft with better trigger and output contract
+- `final/`: final package shape
+- `eval-delta.md`: what changed and why
diff --git a/examples/evolution-frontend-review/eval-delta.md b/examples/evolution-frontend-review/eval-delta.md
new file mode 100644
index 00000000..3b6aa466
--- /dev/null
+++ b/examples/evolution-frontend-review/eval-delta.md
@@ -0,0 +1,19 @@
+# Eval Delta
+
+## v0
+
+- Trigger too vague
+- No output contract
+- High under-trigger risk
+
+## v1
+
+- Trigger became more specific
+- Added review categories
+- Better routing, but output still underspecified
+
+## final
+
+- Stronger trigger coverage
+- Output contract added
+- Review standard became reusable for team workflows
diff --git a/examples/evolution-frontend-review/final/SKILL.md b/examples/evolution-frontend-review/final/SKILL.md
new file mode 100644
index 00000000..850d08cb
--- /dev/null
+++ b/examples/evolution-frontend-review/final/SKILL.md
@@ -0,0 +1,22 @@
+---
+name: frontend-review
+description: Review frontend code for accessibility, risky UI security issues, missing states, and common UX regressions. Use when asked to review React, UI, frontend components, forms, a11y, or pre-merge frontend quality.
+---
+
+# Frontend Review
+
+Review UI code with a pre-merge mindset. Findings come first. Focus on behavior regressions, accessibility gaps, risky form handling, unsafe rendering, and missing user states.
+
+## Workflow
+
+1. Identify the UI surface and user flows affected.
+2. Check accessibility, security-sensitive inputs, loading states, and error states.
+3. Report findings by severity with concrete code references.
+4. Prefer actionable fixes over generic advice.
+
+## Output Contract
+
+- Start with findings, ordered by severity.
+- Include file references for each finding.
+- Call out open questions separately from confirmed issues.
+- If there are no findings, say that explicitly and note residual risks or test gaps.
diff --git a/examples/evolution-frontend-review/raw-workflow.md b/examples/evolution-frontend-review/raw-workflow.md
new file mode 100644
index 00000000..ea5c2e8f
--- /dev/null
+++ b/examples/evolution-frontend-review/raw-workflow.md
@@ -0,0 +1,10 @@
+# Raw Workflow
+
+Review our frontend code before merge for:
+
+- accessibility issues
+- risky HTML rendering
+- missing loading and error states
+- bad form behavior
+
+We want a reusable review skill, not a one-off checklist.
diff --git a/examples/evolution-frontend-review/v0/SKILL.md b/examples/evolution-frontend-review/v0/SKILL.md
new file mode 100644
index 00000000..6e861f04
--- /dev/null
+++ b/examples/evolution-frontend-review/v0/SKILL.md
@@ -0,0 +1,11 @@
+---
+name: frontend-review
+description: Review frontend code before merge.
+---
+
+# Frontend Review
+
+## Workflow
+
+1. Review the code.
+2. Report issues.
diff --git a/examples/evolution-frontend-review/v1/SKILL.md b/examples/evolution-frontend-review/v1/SKILL.md
new file mode 100644
index 00000000..a7efff99
--- /dev/null
+++ b/examples/evolution-frontend-review/v1/SKILL.md
@@ -0,0 +1,14 @@
+---
+name: frontend-review
+description: Review frontend code for accessibility, risky UI security issues, missing loading and error states, and pre-merge regressions. Use when asked to review React, UI, forms, frontend components, or merge-readiness for the frontend.
+---
+
+# Frontend Review
+
+Review UI code with a pre-merge mindset.
+
+## Workflow
+
+1. Identify affected flows.
+2. Check accessibility, rendering safety, form behavior, loading states, and error states.
+3. Report findings by severity with code references.
diff --git a/scripts/diff_eval.py b/scripts/diff_eval.py
new file mode 100644
index 00000000..65bffd00
--- /dev/null
+++ b/scripts/diff_eval.py
@@ -0,0 +1,25 @@
+#!/usr/bin/env python3
+import argparse
+import json
+from pathlib import Path
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Diff two eval reports.")
+    parser.add_argument("before")
+    parser.add_argument("after")
+    args = parser.parse_args()
+
+    before = json.loads(Path(args.before).read_text(encoding="utf-8"))
+    after = json.loads(Path(args.after).read_text(encoding="utf-8"))
+    diff = {
+        "false_positive_delta": after["summary"]["false_positives"] - before["summary"]["false_positives"],
+        "false_negative_delta": after["summary"]["false_negatives"] - before["summary"]["false_negatives"],
+        "precision_delta": round(after["summary"]["average_precision"] - before["summary"]["average_precision"], 3),
+        "recall_delta": round(after["summary"]["average_recall"] - before["summary"]["average_recall"], 3),
+    }
+    print(json.dumps(diff, ensure_ascii=False, indent=2))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/init_skill.py b/scripts/init_skill.py
new file mode 100644
index 00000000..32f41dd8
--- /dev/null
+++ b/scripts/init_skill.py
@@ -0,0 +1,56 @@
+#!/usr/bin/env python3
+import argparse
+from pathlib import Path
+
+
+SKILL_TEMPLATE = """---
+name: {name}
+description: {description}
+---
+
+# {title}
+
+## Workflow
+
+1. Understand the request.
+2. Execute the main task.
+3. Validate the result.
+"""
+
+
+INTERFACE_TEMPLATE = """interface:
+  display_name: "{title}"
+  short_description: "{short_description}"
+  default_prompt: "Use ${name} to ..."
+compatibility:
+  canonical_format: "agent-skills"
+  adapter_targets:
+    - "openai"
+    - "claude"
+    - "generic"
+"""
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Initialize a minimal skill package.")
+    parser.add_argument("name", help="skill folder and frontmatter name")
+    parser.add_argument("--description", default="Describe what the skill does and when to use it.")
+    parser.add_argument("--title", default=None)
+    parser.add_argument("--output-dir", default=".")
+    args = parser.parse_args()
+
+    title = args.title or args.name.replace("-", " ").title()
+    root = Path(args.output_dir).resolve() / args.name
+    (root / "agents").mkdir(parents=True, exist_ok=True)
+    (root / "references").mkdir(exist_ok=True)
+    (root / "scripts").mkdir(exist_ok=True)
+    (root / "SKILL.md").write_text(SKILL_TEMPLATE.format(name=args.name, description=args.description, title=title), encoding="utf-8")
+    (root / "agents" / "interface.yaml").write_text(
+        INTERFACE_TEMPLATE.format(name=args.name, title=title, short_description=args.description[:80]),
+        encoding="utf-8",
+    )
+    print(str(root))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/lint_skill.py b/scripts/lint_skill.py
new file mode 100644
index 00000000..13fcc880
--- /dev/null
+++ b/scripts/lint_skill.py
@@ -0,0 +1,33 @@
+#!/usr/bin/env python3
+import argparse
+import json
+from pathlib import Path
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Lint a skill package for basic structure issues.")
+    parser.add_argument("skill_dir")
+    args = parser.parse_args()
+
+    root = Path(args.skill_dir).resolve()
+    failures = []
+    warnings = []
+
+    if root.name.lower() != root.name:
+        warnings.append("Skill directory is not lowercase.")
+    if " " in root.name:
+        failures.append("Skill directory contains spaces.")
+
+    skill_md = root / "SKILL.md"
+    if skill_md.exists():
+        lines = skill_md.read_text(encoding="utf-8").splitlines()
+        if len(lines) > 300:
+            warnings.append("SKILL.md is getting long; consider moving detail into references/.")
+
+    print(json.dumps({"ok": not failures, "failures": failures, "warnings": warnings}, ensure_ascii=False, indent=2))
+    if failures:
+        raise SystemExit(2)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/run_eval_suite.py b/scripts/run_eval_suite.py
new file mode 100644
index 00000000..2134e6e8
--- /dev/null
+++ b/scripts/run_eval_suite.py
@@ -0,0 +1,67 @@
+#!/usr/bin/env python3
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+
+def run_case(script: Path, description: Path, baseline: Path, cases: Path) -> dict:
+    proc = subprocess.run(
+        [
+            sys.executable,
+            str(script),
+            "--description-file",
+            str(description),
+            "--baseline-description-file",
+            str(baseline),
+            "--cases",
+            str(cases),
+        ],
+        capture_output=True,
+        text=True,
+    )
+    payload = json.loads(proc.stdout)
+    payload["returncode"] = proc.returncode
+    return payload
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Run trigger evaluation across train/dev/holdout suites.")
+    parser.add_argument("--eval-dir", default="evals", help="Root eval directory")
+    parser.add_argument("--description-file", default="evals/improved_description.txt")
+    parser.add_argument("--baseline-description-file", default="evals/baseline_description.txt")
+    args = parser.parse_args()
+
+    root = Path(args.eval_dir).resolve()
+    script = Path("scripts/trigger_eval.py").resolve()
+    description = Path(args.description_file).resolve()
+    baseline = Path(args.baseline_description_file).resolve()
+
+    suites = {}
+    aggregate = {"false_positives": 0, "false_negatives": 0, "precision": [], "recall": []}
+    for name in ("train", "dev", "holdout"):
+        report = run_case(script, description, baseline, root / name / "trigger_cases.json")
+        suites[name] = report
+        aggregate["false_positives"] += report["false_positives"]
+        aggregate["false_negatives"] += report["false_negatives"]
+        if report["precision"] is not None:
+            aggregate["precision"].append(report["precision"])
+        if report["recall"] is not None:
+            aggregate["recall"].append(report["recall"])
+
+    summary = {
+        "suite_count": len(suites),
+        "false_positives": aggregate["false_positives"],
+        "false_negatives": aggregate["false_negatives"],
+        "average_precision": round(sum(aggregate["precision"]) / len(aggregate["precision"]), 3) if aggregate["precision"] else None,
+        "average_recall": round(sum(aggregate["recall"]) / len(aggregate["recall"]), 3) if aggregate["recall"] else None,
+    }
+    output = {"summary": summary, "suites": suites}
+    print(json.dumps(output, ensure_ascii=False, indent=2))
+    if summary["false_positives"] > 0 or summary["false_negatives"] > 0:
+        raise SystemExit(2)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/scripts/validate_skill.py b/scripts/validate_skill.py
new file mode 100644
index 00000000..6ef9e1c7
--- /dev/null
+++ b/scripts/validate_skill.py
@@ -0,0 +1,47 @@
+#!/usr/bin/env python3
+import argparse
+import json
+from pathlib import Path
+import yaml
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(description="Validate a skill package.")
+    parser.add_argument("skill_dir")
+    args = parser.parse_args()
+
+    root = Path(args.skill_dir).resolve()
+    failures = []
+    skill_md = root / "SKILL.md"
+    interface = root / "agents" / "interface.yaml"
+
+    if not skill_md.exists():
+        failures.append("Missing SKILL.md")
+    if not interface.exists():
+        failures.append("Missing agents/interface.yaml")
+
+    if skill_md.exists():
+        text = skill_md.read_text(encoding="utf-8")
+        if not text.startswith("---"):
+            failures.append("SKILL.md missing frontmatter")
+        else:
+            parts = text.split("---", 2)
+            data = yaml.safe_load(parts[1]) or {}
+            for field in ("name", "description"):
+                if not data.get(field):
+                    failures.append(f"Missing frontmatter field: {field}")
+
+    if interface.exists():
+        data = yaml.safe_load(interface.read_text(encoding="utf-8")) or {}
+        meta = data.get("interface", {})
+        for field in ("display_name", "short_description", "default_prompt"):
+            if not meta.get(field):
+                failures.append(f"Missing interface field: {field}")
+
+    print(json.dumps({"ok": not failures, "failures": failures}, ensure_ascii=False, indent=2))
+    if failures:
+        raise SystemExit(2)
+
+
+if __name__ == "__main__":
+    main()