owndiff-pre-commit

OwnDiff Pre-Commit

Use this skill to produce a verified commit candidate. Stop before committing or pushing unless the user explicitly authorizes that Git action.

Workflow

1. Read the latest user request and git status --short. Treat existing worktree changes as user work; never discard them.
2. Inspect the complete staged and unstaged diff. Separate release changes from ignored proof, generated artifacts, caches, research, and unrelated edits.
3. Run the deterministic audit from the repository root with the repository interpreter:

.venv/bin/python skills/owndiff-pre-commit/scripts/audit.py --repo . --staged --run-checks

If nothing is staged yet, omit --staged for the first diagnostic pass. Stage only confirmed release files, then rerun with --staged.

1. Run a leak audit before any commit, push, tag, release, or marketplace/package update:

.venv/bin/python skills/owndiff-pre-commit/scripts/audit.py --repo . --history

The leak audit must cover the current tracked tree, staged diff, unstaged diff, untracked release candidates, and Git history reachable from all local refs. High-confidence private paths or credential patterns are blockers. Redaction markers and secret-adjacent examples are warnings; inspect their context without printing matched text, then remove accidental leaks or document why the warning is an intentional test, security rule, or sanitization example.

1. For release work, verify binary integrity before continuing:
   • Run python scripts/verify_release_assets.py --release-dir dist after local release assets are built.
   • After publishing a tag, run python scripts/verify_release_assets.py --repo owndiff/own-your-diff --tag <tag>.
   • Treat a missing binary, missing .sha256 sidecar, invalid checksum file, or checksum mismatch as a release blocker.
   • The deterministic audit also checks that CI/release workflows keep this verification wired in.
2. Verify documentation semantically:
   • Map behavior claims to current code, config, tests, or observed command output.
   • Verify unstable install syntax with the installed CLI's --help first.
   • When local help is unavailable, use only the agent vendor's official documentation and cite it in the work log.
   • Do not claim an agent, command, browser review, or workflow was tested unless it actually ran.
   • Remove a document or asset only when it is duplicated, stale, unreferenced, generated, or outside the product's supported workflow. Preserve README.md, skills/owndiff/SKILL.md, SECURITY.md, CONTRIBUTING.md, licenses, manifests, and referenced media unless evidence shows they are obsolete.
3. Run OwnDiff on the exact candidate diff. For medium/high/critical risk, use the active agent model to generate every question and answer choice, validate the response, and complete the real browser review gate.
4. Re-run the audit, tests, skill/plugin validators, release-asset verification when applicable, and leak audit after every correction.
5. Report:
   • changed and removed files;
   • documentation claims verified and their evidence;
   • test and validator results;
   • OwnDiff gate status and attempt summary;
   • any residual risk or unavailable agent runtime.