Brooks-Lint

Highlights

brooks-sweep has been rewritten from a single-pass unified scan into a 9-step sequential autonomous audit-and-fix loop. After one up-front consent, the pipeline scans, fixes, and verifies four quality dimensions in order — then iterates until the codebase converges or a cap is hit.

The input contract and report format are unchanged. Existing /brooks-sweep invocations continue to work; you'll see more automation, an iteration history, and a cleaner residual section in the report.

What changed

🔄 Sequential four-dimension pipeline

Instead of one merged scan with late-stage classification, sweep now walks four dimensions in a fixed order — each scanning, classifying, applying Safe + Extended-Safe fixes, and verifying via the project test command before the next starts:

Step 0  Pre-flight consent gate
Step 1  Scope enumeration + state init
Step 2  brooks-review pass  (R1–R6 code decay)
Step 3  brooks-test pass    (T1–T6 test decay)
Step 4  brooks-debt pass    (accumulated debt)
Step 5  brooks-audit pass   (architecture integrity)
Step 6  Iteration loop
Step 7  Residual aggregation
Step 8  Sweep report

🛠️ New Fix-Class taxonomy

| Old | New | Behavior |
|-----|-----|----------|
| Auto | Safe | Auto-applies (single-file, fully local) |
| Confirm | Extended-Safe | Now auto-applies when tests + no public-API break (was: always prompted) |
| Manual | Residual | Recorded in report for human review (not applied) |

No mid-pipeline Hard-Confirm. The only user interaction is Step 0 consent.

♻️ Convergent iteration loop

• Re-scans modified files + same-module + static consumers after each round
• Retires findings that fail 3 fix attempts to an unresolvable set (never re-queued)
• Caps non-critical rounds at 3 (configurable via .brooks-lint.yaml sweep.max_iterations)
• Critical findings iterate until resolved OR retired

📝 Report additions

• Per-dimension fix summary (Safe / Extended / Reverted / Residual)
• Iteration history (round-by-round classification)
• Fix Log with explicit applied / reverted / retired outcomes
• Residual items annotated with the reason they were not applied

Design inspiration

Autonomy controls (3-retry retirement, round-cap escalation, consent gate) are adapted from logic-lens /logic-fix-all's proven pipeline. brooks-sweep applies the same guardrails to the Brooks-Lint book-grounded R/T risk framework.

Upgrade

No breaking changes to inputs or report structure. Just /plugin marketplace update then /plugin install brooks-lint@brooks-lint-marketplace, or update your installed skills:

cp -r skills/* ~/.claude/skills/brooks-lint/

Full changelog: CHANGELOG.md

🤖 Co-authored with Claude Opus 4.7 (1M context).