Use when code changes are ready for quality review, after implementation completes, or before committing — 'review this', 'check quality', 'what needs fixing'.
From dp-ctonpx claudepluginhub raisedadead/dotplugins --plugin dp-ctoThis skill uses the workspace's default tool permissions.
references/review-full.mdreferences/review-pushback.mdreferences/review-quick.mdreferences/review-sweep.mdDesigns and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Implements structured self-debugging workflow for AI agent failures: capture errors, diagnose patterns like loops or context overflow, apply contained recoveries, and generate introspection reports.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
Detect the current dp-cto stage to infer the appropriate review mode:
bash -c 'source ${CLAUDE_PLUGIN_ROOT}/lib/dp-beads.sh && dp_beads_active_epic_and_stage 2>/dev/null'
Returns stage\tepic_id. If it fails or returns empty, treat as idle.
Stage → mode mapping:
| Stage | Mode | Rationale |
|---|---|---|
planned | pushback | Plan just completed — challenge it |
running or reviewing | full | Code changes to review |
complete | sweep | Final cleanup pass |
idle / suspended / unknown | — | Present menu (see below) |
When stage is ambiguous or idle, use AskUserQuestion with options: Quick / Full / Sweep / Pushback.
When args are non-empty (internal invocation from orchestrator), use the args as the mode hint directly (e.g., args pushback → pushback mode).
Read ${CLAUDE_PLUGIN_ROOT}/shared/cto-boundaries.md
Read ${CLAUDE_PLUGIN_ROOT}/shared/evidence-grounding.md
Based on the selected mode, load the corresponding reference file:
${CLAUDE_SKILL_DIR}/references/review-quick.md${CLAUDE_SKILL_DIR}/references/review-full.md${CLAUDE_SKILL_DIR}/references/review-sweep.md${CLAUDE_SKILL_DIR}/references/review-pushback.mdFollow the protocol described in the loaded reference file.
Quick -- Single-lens focused review. Pick one lens (security, simplification, test gaps, linting, performance, or docs). Run one reviewer, collect findings, triage, and present.
Full -- Multi-perspective review across 2-3 parallel lenses (primary + adversarial). Collect findings from both, deduplicate by file:line, consolidate into a single report, and run fix rounds for CRITICAL and WARNING items.
Sweep -- Entropy management scan. Detect drift across dead code, inconsistent patterns, stale comments, and naming violations. Fix CRITICAL and WARNING drift; report SUGGESTION items.
Pushback -- Adversarial challenge of proposals, plans, or recommendations. Assess scope validity, feasibility risks, and evidence gaps. Verify claims before accepting.