| name | clankstamp |
| description | Create a clankstamp replay artifact (a "stamp") after meaningful code changes so a human can replay what changed, why it changed, where it changed, and what needs review. Trigger when modifying code, config, tests, migrations, dependencies, or architecture. |
clankstamp Skill (Claude Code)
This is the Claude Code variant of the clankstamp skill. Behavior is identical to the open SKILL.md standard variant — see the rules and recommended flow below.
When to use
Trigger this skill when you make meaningful changes to:
- Source code (functions, classes, modules)
- Configuration files (auth, routing, feature flags)
- Database migrations
- Tests (added, removed, or changed)
- Dependencies (added, upgraded, removed)
- Architecture decisions (new services, refactors)
Skip for: typo fixes, comment-only changes, formatting-only changes.
Rules
- Do not store hidden chain-of-thought.
- Do store concise rationale, evidence, assumptions, risks, and review questions.
- Prefer exact file paths, symbols, hunks, and test evidence.
- Order tour steps by human comprehension, not file path.
- Keep each step short.
- Highlight risky changes: auth, billing, permissions, migrations, destructive operations, dependencies, config.
- Always note whether tests were added or run.
- Always include git base/head metadata when available.
- Always write JSONL artifacts under
.clankstamp/.
Recommended flow
clankstamp start --title "<task title>" --task "<task id>"
clankstamp record file.read <path>
clankstamp record file.write <path>
clankstamp record test.run -- <test command>
clankstamp finalize --auto-tour
clankstamp create --from-worktree --title "<task title>" --auto-tour
Tour step requirements
Each step answers: what changed, why, where, what evidence, what to review.