← ClaudeAtlas

perfectlisted

Post-implementation quality pass: evaluate, grill loop, verification squad, and proof of correctness with user gates. Phase /4 of the sprint pipeline. Also invoked as /4.
turqoisehex/cc-sentinel · ★ 4 · AI & Automation · score 77
Install: claude install-skill turqoisehex/cc-sentinel
# /perfect — Post-Implementation Quality Pass (alias: /4) `/perfect` (session files) or `/perfect <subsystem>` (named target) **Channel:** CT=`CURRENT_TASK_chN.md` (channeled) or `CURRENT_TASK.md`. Scripts: `SENTINEL_CHANNEL=N`. `[chN/]`=dispatch subdir, `[_chN]`=file suffix, `[chN_]`=squad prefix. Full rules: `.claude/reference/channel-routing.md`. **Step 0:** Before any other work, TaskCreate every step. Mark in_progress->completed. ## Delegation **Default mode:** Steps marked DELEGATE: spawn Sonnet subagent via `Agent(model: "sonnet")` with the delegation prompt. Output to `verification_findings/` paths specified per step. **Duo mode:** Steps marked DELEGATE: update CT first, write self-contained prompt to `verification_findings/_pending_sonnet/[chN/]`, wait via `bash ~/.claude/scripts/wait_for_results.sh <paths>`. ## Phase 1: Scope and Evaluate ### 1. Scope - Bare: session files, fall back to `git diff main...HEAD`. - Named subsystem: read every file -> `verification_findings/perfect_inventory[_chN].md`. ### 2. Evaluate Read in-scope files + authoritative spec + project rules. Catalog: accidental complexity, incomplete migrations, over/under-engineering, naming lies. Write to `verification_findings/perfect_evaluation[_chN].md`. ### 3. Branch — user gate - **Already elegant** -> Phase 2. - **Sound approach, messy execution** -> Step 4, then Phase 2. - **Mediocre approach** -> Step 5, then Phase 2. Present assessment. Wait for approval. ### 4. Simplify DELEG