audit

# procoder:audit The engine sweeps; you rank. A repo has too many findings to list — the job is to say which three mistakes are being repeated. ## Procedure 1. **Sweep with the engine.** `node <plugin>/bin/procoder.js check .` (or the path in `$ARGUMENTS`). These findings are deterministic and already reviewed. Report them as-is. Never re-derive them by reading files, and never drop one because it looks minor. 2. **Group and count.** By rung, then by directory. A rung's worst directory is the one with the most findings, not the most files. 3. **Sample-read the three worst files** for what the engine cannot compute: rung-3 naming quality, rung-4 semantics (is this symbol genuinely unreachable?), rung-1 trust-boundary reasoning. Name a line for each judgment finding or drop it. 4. **Rank.** Order findings SAFE → TRUE → OBVIOUS → ALONE, then by count of the same rule. Cap the individual list at the **top 15**. 5. **Name the top three systemic patterns.** One repeated mistake matters more than thirty instances of it. For each, give the single change that fixes the whole class. ## Output First a rung summary table: | Rung | Count | Worst directory | |---|---|---| | 1 SAFE | 12 | `api/` | | 2 TRUE | 31 | `workers/` | | 3 OBVIOUS | 84 | `legacy/` | | 4 ALONE | 22 | `legacy/` | Then the ranked top 15, one line per finding: ``` [1 SAFE] api/users.ts:42 raw req.body.role into authz check → validate + server-side role lookup [4 ALONE] legacy/...