slop-review

Files (1)

• SKILL.md

SKILL.md (first lines)

# Slop review

## Checklist (after `slop-scan.mjs` and re-running gates)
- **Hardcoding:** any branch on literal test values? Functions returning constants? Tests that only pass because of the fixture?
- **Static data:** sample arrays in components/handlers/services instead of real sources.
- **Tests:** do they fail if you break the implementation? Spot-check one: mutate a line, re-run, expect red. Mocks of the unit under test = fake test.
- **Claims vs reality:** every "PASS" in the report reproduced by you.
- **Scope:** files touched outside the brief; new dependencies; config edits.
- **Structure:** duplicated blocks, god files (>300 lines), new helper that already existed, `any`.
- **UI:** generic AI look (purple gradients, emoji icons, centered-everything hero), missing empty/loading/error states, broken at 375px, ignoring existing tokens.

## Verdict
- Clean → land.
- Minor (≤3 defects) → delta brief with the exact defects, same tier, `--resume-last`.
- Structural/hardcoded/faked → discard (`git restore -SW . && git clean -fd` on the implementer's files only, after checking `git status`), re-brief one tier up.

## Incident loop (do this every time a pattern shows up)
1. Append to `~/orchestra/guardrails/INCIDENTS.md`:
   `- YYYY-MM-DD | agent/model | pattern-id | one-line what happened | fix applied`
2. `node ~/orchestra/bin/orch.mjs log --type incident --agent agy --model <m> --task "<t>" --summary "<pattern>"` (shows red on the dashboard).
3. If the same pattern appears **twice**: add one sharp rule to `guardrails/gemini-rules.md` (or `codex-rules.md`) — imperative, concrete, with a forbidden example. Keep the file under ~60 lines; merge rather than append.
4. If a regex could catch it, add a rule to `guardrails/slop-scan.mjs`.
5. If a whole class of task keeps failing on agy, record it in the orchestrator skill's "Never give agy" line.

From ahmedashraf0001/claude-orchestra, MIT.

More from ahmedashraf0001/claude-orchestra