Definition Of Done

Definition of Done

Doctrine

The order is load-bearing. The integrity manifest must hash the settled tree, so asset-integrity:write always runs last, on its own, from the repo root — never folded into a parallel generator batch, never run from a subdirectory.

Trigger

• Before declaring any change in this repo finished.
• Whenever the stop hook reports uncommitted or untracked changes.

Input

The list of paths you touched in this change.

Decision matrix

Pick the steps that apply based on what changed — most changes trigger more than one row.

• catalog / agents / skills / roles / providers changed → npm run manifest:write:all,

then re-run asset-integrity LAST on its own (the parallel-generator ordering caveat in CLAUDE.md).

• skills/** changed → npm run manifest:write.
• catalog/model-policy.json or model-registry.json changed → `npm run

model-policy:check (and model-policy:apply` if the projection changed).

• tools/vfa-tui/** changed → `cd tools/vfa-tui && cargo fmt --check && cargo clippy

--all-targets -- -D warnings && cargo test. Return to the repo root afterwards — a persisted cd` has broken integrity runs before.

• any root file / agents/ / plugins/ / package.json changed → asset-integrity

refresh required.

The invariant sequence

Always, in this order:

1. Applicable generators from the decision matrix above.
2. python3 tests/validate-asset-integrity.py --write from the REPO ROOT, last, on its own.
3. npm run validate — zero failures.
4. npm run lint:spell and npx --yes markdownlint-cli2 "**/*.md" "#node_modules" — zero

failures.

1. git status clean after committing with a scoped conventional-commit message.
2. git push -u origin (retry with backoff on network errors only).

Footguns

• grep -cE 'FAIL|ERROR' exits 1 on a count of 0 — do not chain it with && before the

commit, or a clean gate run looks like a failure and blocks you.

• Run integrity from the repo root — a leftover cd into tools/vfa-tui makes the write

silently target a nonexistent path.

• Warnings from model-policy are exit-0 by design — read them, don't treat them as failures.
• Never edit catalog/asset-integrity.json by hand.

Output

The checklist above with pass evidence per step, then the commit hash and push confirmation.

Delegation note

Gate runs may be delegated to a Haiku subagent per agentic-delegation's "Gate run" workflow template, but the orchestrator reads the results and owns the commit — a delegate's self-report that gates passed is not verification.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: VincentChuWaiChow
• Source: VincentChuWaiChow/vanguard-frontier-agentic
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.