master-optimize

name	master-optimize
description	Audit an authorized local repo or feature for wasted memory, CPU, rendering, network, disk, battery, and unnecessary complexity. Use for Master-Optimize, performance review, resource audits, leak hunts, rendering bloat, or lean-code opportunities. Produces an evidence-backed analysis report with measurement receipts and risk-triggered independent review; it never changes project code.

Master-Optimize

Produce a source-grounded resource-efficiency report. Do not implement fixes in this run.

Core Rules

• Cite file:line, the waste mechanism, path temperature, user impact, confidence, and a leaner alternative for every finding.
• Label evidence as measured, read in source, or hypothesis. Never present estimates as measurements.
• Capture execution receipts for benchmarks or measurement commands and compare with the last trustworthy baseline/report when available.
• Read the optimize log first; do not re-report fixed/dismissed findings without new evidence.
• Use full or focus mode. Periodic full audits may narrow unchanged clean areas using the log.
• Independently verify every cited finding yourself. Trigger one fresh-context reviewer only when the report contains a High finding, a critical security/auth/data/execution/release boundary, a broad architectural recommendation, or a high-impact claim dependent on nontrivial measurement.
• Ordinary Medium/Low focus reports may finish self-verified. Record why independent review was used or skipped; never imply review occurred.
• Verify reviewer additions against source before accepting them. Reviewer advice is not authority.
• Write reports under ~/.auto-master/artifacts/<project-slug>/optimize/.
• Preserve unrelated work, do not expose secrets, and never automatically delete or refactor anything.

Pipeline

1. Startup - confirm repo/branch, choose full/focus, read the log, and identify the prior baseline.
2. Map - use auto-research or a lean inline map for architecture, hot paths, resource lifecycles, and existing measurement tools.
3. Scan - apply references/finding-categories.md; read actual code and record checked-clean areas.
4. Measure - run only cheap relevant repo-native measurements and record receipts. After the same measurement failure twice, diagnose rather than retry blindly.
5. Draft - rank findings by user impact and confidence using references/report-format.md.
6. Review decision - self-verify citations; invoke the selective review in references/peer-review.md only at a defined risk trigger.
7. Correct - confirm reviewer corrections/additions in source; record disputes honestly.
8. Present - save the final or explicitly unreviewed draft report and summarize only findings already in scope.
9. Log - append findings, dismissed claims, checked-clean areas, receipts, baseline delta, and review mode.

Hand approved behavior-preserving fixes to master-refactor; hand behavior changes to the Auto pipeline as a separate authorized run.

References

• Read references/workflow.md before auditing.
• Read references/finding-categories.md before scanning.
• Read references/report-format.md before drafting.
• Read references/peer-review.md when deciding or running independent review.
• Read references/optimize-log.md before log access.