Skill: Scoped Re-audit

<!-- architectural-context Loaded by: commands/review.md (orchestrator) at Phase 2, Pass 2+ only — not on Pass 1. Consumer: the orchestrator reads this file and applies the mapping table manually; there is no programmatic parsing of this file's output. Output contract: the orchestrator extracts the Track A / Track B / Track C / Track D determination from the File-to-Track Mapping section. Track letters (A, B, C, D) must match the agent names used in review.md Phase 2's spawn instructions. Autonomous mode addition: the regression-guard agent is an exception to the A/B/C mapping. It always runs on Pass 2+ in autonomous mode regardless of which files changed. Its spawn decision is not derived from this skill's track table — see "Regression Guard Exception" below. Cross-file dependency: if the mapping table changes, review.md Phase 2 instructions must be re-read to verify consistency with the new track scope definitions. What breaks if format changes: the orchestrator reading this skill expects the mapping to cover commands/, agents/, skills/, hooks/, scripts/, src/, templates/ as patterns. Adding new plugin component types requires a corresponding mapping entry here. -->

Determines which analysis tracks need re-running based on files modified in the previous pass. The orchestrator consults this skill before spawning subagents on Pass 2+.

Trigger

The orchestrator has completed a batch of implementation changes and needs to determine which subagents to spawn for the re-audit pass.

File-to-Track Mapping

Track A (Principles Analyst) is affected when any of these were modified: commands/*.md, agents/*.md, skills/*/SKILL.md, hooks/hooks.json, scripts/*.sh, .mcp.json, src/**, templates/*.md, or the ## Principles section of README.md.

Track B (UX Analyst) is affected when any of these were modified: commands/*.md, agents/*.md, templates/pass-report.md, templates/final-report.md, scripts/*.sh (if they produce stdout), or any file in src/ that produces formatted output or handles errors.

Track C (Docs Analyst) is affected when any file was modified. Documentation drift can be introduced by any change. Track C always runs on re-audit.

Track D (Efficiency Analyst) is affected when any file was modified. Track D evaluates context efficiency patterns that appear in all component types; any component modification may introduce efficiency regressions.

Decision Logic

For each modified file, match against Track A and Track B patterns. Add "C" unconditionally. Add "D" unconditionally (Track D covers all component types and state files). Deduplicate and return the affected tracks.

Conservative Default

If uncertain whether a file change affects a track, include it. The cost of an unnecessary subagent spawn is much lower than missing a regression.

Unchanged Finding Carryover

For tracks NOT re-audited, carry forward all findings from the previous pass as "Unchanged from Pass N." For tracks that ARE re-audited, the subagent produces a fresh analysis that the orchestrator compares against the previous pass to detect status changes, new findings, and resolutions.

Special Case: README.md or docs/DESIGN.md Modified

Track C must re-analyze for internal consistency. Track A should be included if the ## Principles section was modified, since principle definitions are what Track A measures against. Track B is typically unaffected by documentation changes.

Regression Guard Exception [AUTONOMOUS MODE ONLY]

The regression-guard agent (agents/regression-guard.md) is not part of the A/B/C track system. It operates on a fixed spawn rule:

• Always spawned on Pass 2+ in autonomous mode, regardless of which files changed
• Never spawned on Pass 1 (no previously-fixed findings exist)
• Never spawned in interactive mode (regression guard is autonomous-only)

The regression guard does not have a track mapping because it checks previously-fixed findings across all tracks. Its input is the fixed_findings array from .claude/state/plugin-review-writes.json, not a list of modified files. Consult this exception rule before spawning to ensure the regression guard is not incorrectly treated as a track-conditional agent.