← ClaudeAtlas

review-execution-packetlisted

Independently audit a packet claimed as executed or closed - verify the claimed integration, evidence, and closeout against fresh repository/tool state rather than trusting the closeout narrative. Use for a post-hoc audit, not while executing the packet yourself.
mike-arbuzov365/engineering-intelligence-framework · ★ 3 · Code & Development · score 72
Install: claude install-skill mike-arbuzov365/engineering-intelligence-framework
# Skill: Review Execution Packet <!-- Knowledge source: GENERALIZE of the private EI's review-execution-packet SKILL.md. Thin pointer to playbooks/execution-packet-review.md - one canonical source per D-007. --> Full workflow: [`playbooks/execution-packet-review.md`](../../playbooks/execution-packet-review.md). ## Process 1. **Fetch fresh state first, always** (`git fetch --all --prune`). Never reason about merge/integration state from a local ref alone. 2. Build an independent plan-vs-done matrix from the roadmap and session launch files - one row per exit criterion, with the actual evidence and result, not the closeout's own claim. 3. Verify every "merged"/"done" line against fetched state and the relevant hosted-platform query (e.g. `gh pr view`) - not a commit-subject grep (squash/rebase merges don't preserve it). 4. If the packet spans more than one project/repository, confirm a shared contract is present on the fetched state of *every* side. 5. Check that the closeout separates implementation-done from owner-gated/pending, and that any status artifact reflects verified state. 6. Run a normal code-quality review against the correct diff range. ## Output ```markdown Verdict: closed-correctly | gaps-found | false-closeout Findings (one per issue, with evidence and severity): - [HIGH|MEDIUM|LOW] <finding> - <evidence> Fix report: <concrete next steps, referencing playbooks/execution-packet-execution.md for how to close correctly> ```