review-all

name	review-all
description	Orchestrator that runs the full review-* family (agents, skills, rules, docs, memory) and aggregates findings under a single board parent task. Produces a prioritized remediation roadmap. Use for full system audits.
disable-model-invocation	true
user-invocable	true

Full Review Orchestrator

Runs the complete review-* family in sequence and aggregates findings. Use for comprehensive system audits (e.g., post-model-upgrade, quarterly health check, pre-major-initiative sanity check).

Child skills:

1. /review-agents — .claude/agents/*.md
2. /review-skills — .claude/skills/*/SKILL.md
3. /review-rules — CLAUDE.md + .claude/rules/*.md
4. /review-docs — docs/**/*.md
5. /review-memory — ~/.claude/projects/<slug>/memory/*.md
6. /review-board — Vibe Board state (Firestore via mcp__agent-board__*)
7. /review-security — .claude/ configuration security posture

Cites ../_shared/review-checklist.md and ../_shared/anthropic-configuration-guide.md.

Owner: docs-manager agent.

When to invoke

• After a Claude model upgrade (e.g., Opus 4.7 → 4.8)
• After major infrastructure migrations (e.g., Supabase → MIG)
• Quarterly (every ~3 months)
• Before a major initiative where the .claude/ config needs to be trusted

Output contract

A single aggregate report posted to the board + a lean summary to the main agent.

---

Step 1 — Create audit parent

On the current board project (or create a new "System Audit YYYY-MM-DD" project), create a parent task:

[review-all] System Audit <date>
  status: in_progress
  priority: high
  description: Orchestrated run of review-agents, review-skills, review-rules, review-docs, review-memory.

Capture the parent task ID for all child skills to reference.

Step 2 — Run child skills in sequence

Run in this order (sequential — rules first so deprecated-reference findings surface once centrally, not N times across child reviews):

1. /review-rules — core instructions are the foundation
2. /review-agents — agents are the most consequential artifacts
3. /review-skills — skills reference agents and rules
4. /review-docs — docs may reference agents, skills, rules
5. /review-memory — memory may reference anything
6. /review-board — Vibe Board hygiene, independent of config layer
7. /review-security — runs last so it sees all flagged references and can cross-check for secrets in files flagged for other reasons

For each child, pass the parent task ID so subtasks roll up under it.

Step 3 — Aggregate findings

After all child skills finish, query the board for subtasks rolling up to the audit parent. Group by:

• Severity (critical / high / medium / low)
• Domain (agents / skills / rules / docs / memory)
• Theme (model tier, staleness, frontmatter, cross-reference, duplication, gap)

Step 4 — Cross-cutting findings

Look for patterns that only emerge when viewing all findings together:

• Model rubric drift across multiple agents (most are on Opus; should audit globally)
• Deprecated-reference cluster (same retired service mentioned in multiple places)
• Rule-hook misalignment that shows up in both rules and agent behavior
• Naming inconsistencies across agents/skills (e.g., nextgen-* vs next-gen-*)
• Documentation index drift showing up in review-docs, but the underlying cause is missed updates during agent rewrites

Promote cross-cutting patterns to critical priority even if individual findings were lower severity.

Step 5 — Remediation roadmap

Under the audit parent, create ordered child tasks representing the suggested fix order. Priority:

1. Critical — broken behavior, security, production risk
2. High — wrong model tier on critical-path agents, stale references that cause wrong behavior
3. Cross-cutting patterns elevated here
4. Medium — clarity, redundancy, style
5. Low — polish

Include blast-radius annotation: how many other artifacts reference this one?

Step 6 — Summary

Return to main agent (≤ 150 words):

SYSTEM AUDIT COMPLETE — <date>

Scope:
- [N_a] agents, [N_s] skills, [N_r] rules, [N_d] docs, [N_m] memory files reviewed

Findings (total: [T]):
- Critical: [C]
- High: [H]
- Medium: [M]
- Low: [L]

Cross-cutting themes:
- [theme 1]: [count]
- [theme 2]: [count]

Model rubric drift: [X] agents flagged for reassignment
Deprecated references: [Y] occurrences across [Z] files

Board parent task: <id>
Next recommended action: [specific first fix]

Anti-patterns for the orchestrator itself

• Do NOT dump all child-skill findings into the conversation (they live on the board)
• Do NOT run children in parallel — some depend on others' outputs
• Do NOT skip creating the aggregate parent — individual child findings without a roll-up are lost
• Do NOT elevate every finding to critical — discipline the tiers

Tools used

• Skill tool (to invoke each child skill)
• Board MCP (create_task, update_task, log_activity, get_tasks)