# Stip Check

> Reconcile an applied change with every acceptance criterion using fresh, inspectable evidence.

- Skill: `bryann2k/stip-check` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add bryann2k/stip-check`
- Raw SKILL.md: https://api.skillmd.com/api/skills/bryann2k/stip-check/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: Apache-2.0
- Author: BRYANN2K (https://skillmd.com/u/bryann2k)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/bryann2k/stip-check

---


# stip-check

Read the approved contract and relevant extension check references. This stage evaluates the result; it does not redesign the feature or silently change requirements.

For a schema-v2 change, read [references/orchestration.md](references/orchestration.md). Dispatch the planned check tasks to native sessions distinct from their implementers, using configured verification/security profiles or inherited defaults. Identify a stable integrated candidate and ask for concrete defects and missing observations without prescribing a verdict. The coordinator reconciles their results and accepts contributions only after inspection. All apply/check tasks must be accepted and all workers quiescent before the engine can finalize check.

1. Exercise or inspect the actual behavior required by each criterion. A build proves compilation, not UX, authorization or deployment. Distinguish failures from unavailable checks.
2. Run `snapshot` after the last relevant source change. Create a results JSON outside the repository, or pass it through stdin: `{"subject_digest":"<snapshot digest>","criteria":[{"id":"AC-1","status":"passed","evidence":"<actual command/result or observation locator>"}]}`.
3. Run `check <id> --results <file-or-dash>`. Cover each criterion exactly once. Valid statuses are passed, failed, unverified, not-applicable; only all passed advances to checked. A not-applicable requirement needs a contract correction and reapproval rather than silent acceptance.
4. Report gaps. Corrections go through stip-apply when authorized, then a new check. When all criteria pass, proceed to stip-docs.

The command checks report consistency and working-tree identity. It cannot authenticate evidence or perform semantic verification itself. Never fabricate a report merely to advance state. Do not include credentials or private payloads in evidence.md.

If corrections change the candidate, obtain fresh relevant observations and reconcile the worker records before submitting new evidence. A completed native session, a favorable review or an accepted contribution is not by itself a passed acceptance criterion.

## Runtime

Use the Python runtime bundled with this skill, independent of the current directory:

```text
python3 <this-skill-directory>/scripts/workflow.py --root <physical-project-root> <command>
```

Python 3.10+ and Git are required for implementation and archive. Run `--help` or `<command> --help` for exact arguments. The runtime is offline and never installs tools. System/developer instructions, current user intent and environment permissions remain authoritative.