← ClaudeAtlas

staged-coding-6-gatelisted

staged-coding stage 6 - self-check (the gate). A fresh checker subagent runs a bug hunt, a security review and a simplicity review against the diff until clean, plus a silent-failure sweep and quote-or-demote finding rules. Signs its ledger line on finish. Triggers: /staged-coding-6-gate, or invoked in order by /staged-coding.
Shadysmetools/claude-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill Shadysmetools/claude-skills
# Stage 6 - self-check (the gate) Part of the `/staged-coding` chain. Runs after stage 5 (house style) and before stage 7 (logged-in browser). The conductor carries the cross-cutting rules; read them there. ## Inputs - The full diff since the branch was cut. - The ticket (for spec-compliance: what was actually asked). - Whether the diff touches security-sensitive surface (auth, payments, credentials, data access). ## Who runs it: a FRESH checker, never the builder Stage 6 is a CHECKPOINT, not build work. Spawn a **fresh subagent on `config.checker_model`** (Agent tool, the strongest model you have, maximum effort), handed only the diff and the ticket, never the builder's chat history. It signs with `--by <seat>@<checker_model>`, so judge != author is visible in the ledger and `sign.py check` can verify it. If the checker model is unavailable, use the next strongest and say so in the evidence line. ## What to do All of the following, in order, fixing findings until clean: 1. **Bug hunt** - e.g. `/code-review`, or a bug-radar style skill scaled to the diff's size and risk. 2. **Security review** - `/security-review`; mandatory on any security-touching diff. 3. **Simplicity review** - a YAGNI/simplify pass (e.g. `/simplify`): hunt what can be deleted from your own change. 4. **Silent-failure sweep** - hunt exactly five shapes: - empty catch blocks; - log-and-forget (caught, logged, execution continues as if nothing happened); - dangerous fallbacks (`.catch(() =>