Skills Reference
This page documents all slash commands (skills) available to Claude Code agents working on the longterm-wiki project. Skills are organized into four categories based on their purpose.
Agent Lifecycle
These skills manage the session lifecycle: initialize, work, ship, and reset.
| Command | Description | When to use |
|---|---|---|
/agent-init | Initialize agent checklist, register in DB, establish context | First thing in every session |
/agent-ship | Verify checklist, polish PR, create session log, push | End of every session |
/agent-push-and-verify | Run gate checks, push to GitHub, poll CI until green | Called by /agent-ship; also usable standalone |
/agent-review-pr | Paranoid diff review + execution-based verification | Before shipping non-trivial PRs; auto-triggered by /agent-ship for large PRs |
/agent-next-issue | Pick highest-priority GitHub issue, signal start, begin work | When starting a new issue-driven session |
/agent-reset | Close DB session, reset git to clean main, clean up local files | After shipping a PR to prepare for next task |
Typical session flow: /agent-init → work → /agent-ship → /agent-reset
Maintenance & Health
Periodic tasks that keep the codebase, wiki, and process healthy.
| Command | Description | Cadence |
|---|---|---|
/maintain | Full maintenance sweep: review PRs, triage issues, detect cruft, take action | Weekly (or after batch of PRs merge) |
/maintain-pr-patrol | Scan all open PRs for issues (CI failures, conflicts, bot comments) and fix them | Periodically, or via daemon mode |
/maintain-audit | Strategic codebase health review: complexity trends, dead code, system ROI | Biweekly |
/maintain-retrospective | Analyze recent PR patterns, session logs, process friction | Weekly |
/maintain-qa-sweep | Adversarial QA: break things, find bugs, audit data integrity, file issues | Daily via /loop 24h /maintain-qa-sweep |
Content & Data Review
Deep analysis of specific entities, data quality, and knowledge coverage.
| Command | Description | Input |
|---|---|---|
/review-entity-deep-dive | Comprehensive review of one entity: completeness, accuracy, connections, recency | Entity slug (e.g., anthropic) |
/review-ontology | Analyze entity structure: sub-entity candidates, relationship gaps, statement placement | Entity slug |
/review-knowledge-gap | Identify missing topics, thin coverage areas, relationship gaps | Optional focus area |
/review-tablebase-enrich | Enrich structured data (personnel, funding, benchmarks) via web research | Runs from task queue |
Operations
Standalone operational tasks with their own workflows.
| Command | Description | Schedule |
|---|---|---|
/auto-update | Fetch news, route to wiki pages, make targeted edits | Daily at 06:00 UTC (GitHub Actions) or via subscription |
/maintain (cron) | Maintenance sweep via launchd | Every 4 hours (gated on PR count) |
Naming Convention
Skills follow a prefix-based naming scheme for discoverability:
agent-*— Session lifecycle (init, ship, push, review, reset, next-issue)maintain-*— Periodic health checks and cleanup (audit, retrospective, qa-sweep, pr-patrol)review-*— Deep analysis of specific content/data (entity-deep-dive, ontology, knowledge-gap, tablebase-enrich)auto-updateandmaintain— Top-level operational commands
Skill Locations
Skills are defined in two places:
- Commands (
.claude/commands/*.md): Simple markdown files, one per command - Skills (
.claude/skills/*/SKILL.md): Directory-based skills with frontmatter for tool restrictions
Both are loaded automatically by Claude Code.