weside-ai

ac-review

Checks a branch against its story's acceptance criteria and the repo DoD: one evidence row per AC and DoD item, verdict PASS/BLOCKING. Hunts no bugs. Triggers: "/we:ac-review", "are the ACs met".

weside-ai 0 Updated 1w ago
GitHub

Install

npx skillscat add weside-ai/claude-code-plugin/ac-review

Install via the SkillsCat registry.

SKILL.md

/we:ac-review

An AC/DoD check, not a bug gate: the bench (27.09.2026) measured 12 false blockers when this check
also hunted bugs. Bugs belong to the CI review on the PR (and a local code-review for money, auth, tenant or migration work).
Report a suspected bug in one line outside the verdict; it never blocks here.

  1. Key from the branch (<type>/<KEY>-…) → docs/plans/<KEY>-story.md (## Acceptance Criteria,
    ## Verification, ### Phase N with **Files:**) and the ticket with its comments
    (${CLAUDE_PLUGIN_ROOT}/references/ticketing.md). No plan → check against the ticket and say so.
  2. Diff: git diff $(git merge-base origin/<base> HEAD) plus the uncommitted changes; the base is
    the PR's baseRefName, else the remote's HEAD.
  3. One row per AC: Pass only with a citation (test name, file:line or commit). Is the feature
    reachable from the UI or CLI entry point the AC names?
  4. One row per applicable item of .weside/dod.md when the repo has one; without it, only these
    rows: every planned phase's files changed; verification receipt present (the four labels filled,
    or not-applicable/not-observed with its reason, quoted as evidence); no new TODO/FIXME. An
    item whose evidence cannot exist yet (PR, CI, ticket move) is N/A naming the stage that owes it.
  5. Output a table | Item | Pass/Fail/N/A | Evidence |, then VERDICT: BLOCKING when an AC is unmet
    or a DoD row fails, else VERDICT: PASS. Write no file.