ve-kit

Use when setting up a new project with the ve-kit framework, adding a layer to an existing setup, or upgrading an older .claude/ config. Runs the ve-kit BOOTSTRAP protocol interactively — prerequisite checks, project-info questions, writes all rules/hooks/skills/agents files, builds the Vibe Board MCP server, self-verifies at the end. Triggers on "/bootstrap", "set up this project with ve-kit", "run the bootstrap", or "upgrade my ve-kit setup".

Full autonomous RIPER cycle in one command. Runs RESEARCH → PLAN → REVIEW → EXECUTE → REVIEW with auto-advance on REVIEW PASS. Stops only on REVIEW FAIL, user-required decisions, or safety guards.

Use when entering PLAN mode or when the user says "Enter P". Creates structured implementation plans with Vibe Board tasks and specialist agent assignments for each checklist item.

Review all specialist agent definitions in .claude/agents/ for accuracy, currency, and Anthropic best-practice compliance. Flags stale references, wrong model assignments, missing examples, merge/split candidates, and gap-analysis coverage. Creates board tasks for every finding.

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.

Review the Vibe Board (Firebase Firestore) for stale tasks, orphans, duplicates, done-but-unmarked work, abandoned projects, session handoff quality, and referential integrity. Creates remediation board tasks for every finding.

Review the docs/ hub for broken links, staleness, duplicate coverage, index currency, and size bloat. Creates board tasks for every finding.

Review the per-project auto-memory directory at ~/.claude/projects/<slug>/memory/ for index currency, stale entries, duplicates, type correctness, and uncaptured patterns. Creates board tasks for every finding.

Review the core instruction layer — CLAUDE.md, .claude/rules/*.md, and RIPER/CAT mode definitions — for size, staleness, cross-reference integrity, alignment with active hooks, and redundancy. Creates board tasks for every finding.

Security audit of the .claude/ configuration layer — secrets leaked in agent/skill bodies, overly broad permissions, deny list gaps, hook script shell injection, MCP server tool grants, dangerous agent instructions. Distinct from the built-in /security-review which audits code changes. Creates board tasks for every finding.

Use when entering REVIEW mode or when the user says "Enter RE". Validates plans or implementations by invoking specialist sub-agents based on what changed. Detects whether this is a post-plan review or post-execute review automatically.

Review all skill definitions in .claude/skills/ for frontmatter correctness, progressive-disclosure hygiene, trigger clarity, merge/split candidates, and gap analysis. Creates board tasks for every finding.

Mine recent Vibe Board review findings + worker batch outcomes for RECURRING patterns, then propose preventive guardrails — LESSONS.md rules (applied), agent-definition hardening (proposed), and flagged CLAUDE.md / rules proposals (human approval only). The feedback loop that turns "what reviewers keep catching" into "encoded so it stops happening." Run manually or after every ~5 worker batches.