zeugnis-code-change
Use when the user asks for a pre-commit quality check on staged or uncommitted changes, wants a quick zeugnis pass over 'what I just changed' before committing, or asks 'is this diff okay to commit'. Only runs the domains whose file patterns match the changed files, and always produces an advisory verdict that never blocks the commit.
- License 许可
- MIT
- 最近更新
- 2026-11-02
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- zeugnis-code-change
- description
- Use when the user asks for a pre-commit quality check on staged or uncommitted changes, wants a quick zeugnis pass over 'what I just changed' before committing, or asks 'is this diff okay to commit'. Only runs the domains whose file patterns match the changed files, and always produces an advisory verdict that never blocks the commit.
03 SKILL.md 正文
Run a fast, changed-files-scoped zeugnis pass across whichever domains are
actually relevant to what changed. This is a lighter-weight sibling of the
four full domain-audit skills — each Find-phase pass here is a single,
un-verified check (the code-change verdict is advisory by construction,
so it does not carry the same mandatory-verification guarantees the full
audits do). If the user wants a fully verified audit, tell them to run
the specific zeugnis-*-audit skill instead.
Steps
- Determine
repo_root. Get the changed-file list:
(orgit -C <repo_root> diff --staged --name-onlygit diff HEAD --name-onlyif the user means uncommitted rather than staged changes — ask if ambiguous). Write the list to a scratch JSON file as a plain array of repo-relative paths. - Run:
with NOpython3 -B "${CLAUDE_PLUGIN_ROOT}/scripts/code_change_review.py" <repo_root> \ --changed-files-json <path-from-step-1>--domain-findingsyet, just to see which domains matched. If it exits 1, it printed "zero domains matched" to stderr — relay that to the user verbatim and stop; there is nothing to review. - For each matched domain printed in step 2's
matchedDomains, produce a lightweight findings JSON scoped to only the matched files:dependency_audit: only if a manifest file changed. Runscripts/dependency_audit.py <repo_root> --skip-verification— it performs the real bounded-timeout registry lookup and typosquat heuristic without the mandatory independent re-check, matching this pass's single-check, advisory rigor. Optionally supplement it with thedependency-auditoragent in Find mode for ambiguous names — a supplement to the script's findings, never a replacement for it.assertion_audit: dispatchassertion-auditorin Find mode only, scoped to the changed source/test files.contract_drift: dispatchcontract-auditorin Find mode only, scoped to the changed files.agentic_reliability: dispatchagentic-reliability-auditorin Find mode only, scoped to the changed skill/agent/workflow files. Write each domain's findings to its own scratch JSON file.
- Re-run the script, now passing every matched domain's findings:
(only passpython3 -B "${CLAUDE_PLUGIN_ROOT}/scripts/code_change_review.py" <repo_root> \ --changed-files-json <path-from-step-1> \ --domain-findings dependency_audit=<path> \ --domain-findings assertion_audit=<path> \ ...--domain-findingsfor domains that actually matched). This writesanalysis/zeugnis/reports/CODE_CHANGE_REVIEW.mdwith one section per matched domain (unmatched domains are omitted entirely — never zero-filled) and a verdict line that always readsADVISORY: ...and never suggests blocking anything. - Report the verdict line and per-domain finding counts to the user. Always phrase the summary as advisory — e.g. "N findings worth a look before you commit, but nothing here blocks you."
What NOT to do
- Do not tell the user their commit is "blocked" or "failing" — this skill has no gate, only advice. If the user wants a hard gate, that's a git hook they'd configure themselves, not something this skill does.
- Do not run a domain that didn't match — e.g. don't dispatch
assertion-auditorif no source or test file changed. - Do not invent a zero-finding section for a domain that didn't match, either in the report or in your summary to the user.
04 同仓库更多 Skills
toy-check
A skill that checks things in the repository and reports what it finds.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
demo-check
Reports which generated files changed since the last generation run. Use when the user asks what drifted. Read-only; never edits a file.
zirkel-verify-assumptions
Checks exactly ONE named assumption against real evidence in at most 3 steps and reports one outcome: confidence raised (>=90), reading changed, or still unresolved. Use to resolve a single load-bearing uncertainty — typically a blocking entry surfaced by zirkel-clarify-scope: "verify this one assumption", "is it actually true that X", "check this before we rely on it". For multiple uncertainties, invoke once per uncertainty — the step budget is never shared.