audit

Full-Project Consistency Audit

Orchestrate a comprehensive project audit by spawning a team of parallel research agents, deduplicating their findings, and producing a scored audit report.

Argument Parsing

Parse $ARGUMENTS for these flags:

• --lanes <comma-separated>: Which audit lanes to run. Valid values: infra, frontend, plugins, arch, tests, deps, security. Default: all lanes.
• --previous <path>: Path to a previous audit report. When provided, the writer agent adds Fixed Since Last Audit, New Findings, and Regressions sections.
• --include-stash: When present, also spawn the git-stash-auditor agent.
• --drop-stashes: Used with --include-stash during the fix phase. Tells the ops-cleanup agent to git stash drop all stashes instead of just cataloging them.
• --fix: After the audit completes, automatically run the planning + fix pipeline (Phase 2.5 through Phase 6).
• --fix-only: Skip audit recon entirely. Read the existing docs/AUDIT_FINDINGS.md and go straight to the planning + fix pipeline. Requires a prior audit report to exist.
• --deep: Enhanced audit using installed CLI tools. The default audit uses grep/glob (fast). --deep adds tool-assisted scanning per lane.

Severity Scoring Rubric

All research agents MUST use this rubric when categorizing findings:

• Critical = breaks at runtime OR creates legal liability
• Major = silent failure OR affects >1 subsystem
• Minor = cosmetic, tech debt, no user impact

Output Format for All Research Agents

Each research agent produces a markdown table (max 50 lines) with these columns:

Finding	Severity	Location	Expected	Actual

Rules for all research agents:

• Read selectively -- grep/glob first, then read only relevant lines
• Never read entire files at once
• Keep context lean
• Update your task to completed when done

Team Architecture

Create a team called project-audit. The main agent (team lead) stays lean -- it only spawns agents, routes messages, and summarizes. It never reads files or does research itself.