원클릭으로
skills
List all available oh-my-claude skills with descriptions and when to use each one.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
메뉴
List all available oh-my-claude skills with descriptions and when to use each one.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | skills |
| description | List all available oh-my-claude skills with descriptions and when to use each one. |
Display this list to the user, grouped by user-journey phase. Phase headers are themselves chapter markers — first-time users skim by phase, experienced users grep by skill name.
When you know what you want, jump straight to the verb:
| If you... | Use |
|---|---|
| Want to ship real work end-to-end | /ulw <task> |
| Keep working toward one durable goal across many turns | /goal <objective> |
| Just installed; want to feel the gates fire | /ulw-demo |
| Need to inspect or change configuration | /omc-config |
| Harness feels off / gates stopped firing | /omc-doctor |
| Need a plan before you edit | /plan-hard <task> |
| Need to review existing code | /review-hard [focus] |
| Wonder whether tests still earn their cost | /test-audit [scope] [--apply] |
| Need repo / API context before coding | /research-hard <topic> |
| Have a vague goal — need clarification | /prometheus <goal> |
| Have a draft plan — want it stress-tested | /metis <plan> |
| Stuck debugging — need a second opinion | /oracle <issue> |
| Multiple framings exist — pick consciously | /diverge <task> |
| Need official docs or external API | /librarian <topic> |
| Analyze experimental data / fit with uncertainties | /data-analysis <task> |
| Need verified academic sources / a bibliography | /lit-review <topic> |
| Writing or revising a paper, thesis, or referee response | /manuscript <task> |
| Want an adaptive multi-specialist project evaluation | /council [focus] |
| Want a design-first frontend pass | /frontend-design <task> |
| A gate fired but the work is complete | /ulw-skip <reason> |
| Discovered-scope flagged real items you defer | /mark-deferred <reason> |
| Blocked on a real operational input (credentials, login, dead infra) | /ulw-pause <reason> |
| Rate-limit kill left a resume artifact | /ulw-resume |
| Want session state / counters / flags | /ulw-status |
| Want a time-distribution card | /ulw-time |
| Want cross-session activity digest | /ulw-report |
| Want changelog deltas since your install | /whats-new |
| Want to bootstrap CLAUDE.md for a repo | /atlas |
| Memory directory feels noisy | /memory-audit |
| Want to deactivate ULW mid-session | /ulw-off |
The symptom-table above is a discovery shortcut. Each row maps to a skill in the phase-grouped tables below — those tables carry the full descriptions.
| Skill | Command | When to use |
|---|---|---|
| omc-config | /omc-config [setup|update|change] | Guided multi-choice walkthrough for oh-my-claude.conf — pick a profile (Maximum / Balanced / Minimal) or fine-tune individual clusters. Auto-detects first-time setup vs upgrade vs ad-hoc change. |
| omc-doctor | /omc-doctor | One-shot install health check against the installed ~/.claude tree (core files, @-includes, hook wiring, state root) — from any directory, days after install. Use when the harness feels off or gates stopped firing. |
| ulw-demo | /ulw-demo | Guided 90-second walkthrough that fires the quality gates on a demo file — best first step after install. |
| Skill | Command | When to use |
|---|---|---|
| ulw | /ulw <task> | Any non-trivial task needing full autonomy — coding, writing, research, ops. |
| goal | /goal <objective> | Drive a durable objective relentlessly across turns — re-anchored at every Stop until verifiably achieved or a no-progress stuck-wall (Codex /goal port). Lifecycle: pause/resume/clear/done. |
| plan-hard | /plan-hard <task> | You need a detailed plan before editing — scope decisions, architecture, sequencing. |
| review-hard | /review-hard [focus] | Review existing code or a worktree diff for bugs, quality, missed requirements. |
| research-hard | /research-hard <topic> | Gather repo context, API details, integration points before implementation. |
| frontend-design | /frontend-design <task> | Distinctive, design-first frontend work — establishes palette + typography + layout before code. |
| data-analysis | /data-analysis <task> | Scientific data analysis — explore, fit with honest uncertainties, publication-quality figures with provenance manifests. Peer-review standard, not just "the script ran". |
| lit-review | /lit-review <topic> | Verified academic literature review — every source registry-verified (Crossref/OpenAlex/S2/arXiv); never cites from memory. |
| manuscript | /manuscript <task> | Academic papers, theses, and referee responses — structure → verified citations → draft → rigor critique, in that order. |
| Skill | Command | When to use |
|---|---|---|
| prometheus | /prometheus <goal> | Goal is broad or ambiguous — needs interview-first clarification before planning. |
| metis | /metis <plan> | You have a draft plan and want to stress-test it for hidden risks, missing constraints. |
| oracle | /oracle <issue> | Debugging is hard, root cause unclear, or you need a second opinion on tradeoffs. |
| diverge | /diverge <task> | Multiple credible framings exist — generate 3-5 alternatives BEFORE commit to pick one consciously. Upstream of all the convergent critics. (v1.31.0) |
| librarian | /librarian <topic> | You need official docs, third-party API references, or concrete external examples. |
| ulw-skip | /ulw-skip <reason> | Gate fired but the work is genuinely complete (false positive). One-shot bypass. |
| ulw-correct | /ulw-correct <correction> | Last turn's classification was wrong — record misfire and (when parseable) update intent/domain in-place. Active counterpart to the passive detect_classifier_misfire. (v1.40.x) |
| mark-deferred | /mark-deferred <reason> | Discovered-scope flagged real findings you're consciously NOT shipping this session. |
| ulw-pause | /ulw-pause <reason> | Operational block only — credentials/login, hard external blocker, destructive shared-state action, unfamiliar in-progress state. NOT for taste/policy/credible-approach (v1.40.0: agent owns those under ULW). Cap 2/session. |
| ulw-resume | /ulw-resume [--peek|--list|--dismiss] | Atomically claim a resume_request.json after a rate-limit StopFailure. |
| Skill | Command | When to use |
|---|---|---|
| council | /council [focus] [--deep] | Coverage-map-driven evaluation from the full roster: normally 1–4 primary specialists, optional 0–2 gap-fill, and independent competence-matched verification of up to three findings. --deep escalates selected Sonnet-backed agents only. |
| test-audit | /test-audit [scope] [--apply] | Map behavior contracts to test owners and decide KEEP, EXTEND, MERGE, REPLACE, DELETE, or ADD by unique confidence, runtime, stability, and maintenance cost. Read-only unless --apply is passed. |
| atlas | /atlas [focus] | Bootstrap or refresh CLAUDE.md / .claude/rules for a repository. |
| Skill | Command | When to use |
|---|---|---|
| ulw-status | /ulw-status | Inspect current session state — mode, domain, counters, flags (debugging). |
| ulw-time | /ulw-time [current|last|last-prompt|week|month|all] | Polished time-distribution card — stacked top bar (█ agents · ▒ tools · ░ idle), per-bucket chart, one-line insight. |
| ulw-report | /ulw-report [last|week|month|all] | Markdown digest of cross-session activity — gate fires, reviewers, misfires, Serendipity catches. |
| memory-audit | /memory-audit [--memory-dir <path>] | Classify MEMORY.md entries (load-bearing, archival, superseded, drifted). Read-only. |
| ulw-off | /ulw-off | Deactivate ultrawork mode mid-session without ending the conversation. |
| skills | /skills | Show this list. |
These two skills auto-fire when Claude detects the matching work — there is no slash command to invoke them. They are listed here for discoverability; you never call them directly.
| Skill | Fires on | What it does |
|---|---|---|
| swiftui-pro | SwiftUI code in view | Reviews Swift/SwiftUI for modern-API usage, accessibility, data flow, navigation, performance, and Swift idioms. Partial-loads topic-scoped references. |
| gamedev | Unity / Godot / web-engine game work | Engine-idiomatic review + the frame-grounded run→capture→evaluate→fix loop for Unity (C#), Godot (GDScript/C#), and web engines (Phaser / Babylon / PixiJS / Three.js). |
/omc-config for the guided multi-choice configuration walkthrough, then /ulw-demo to see the quality gates in action./omc-config (auto-detects mode)./ulw (aliases: /autowork, /ultrawork also work)./plan-hard. If the goal is vague, use /prometheus instead./metis./council. It selects only the coverage the task needs; it also auto-triggers under /ulw for broad prompts like "evaluate my project," not a focused feature or subsystem audit./oracle./librarian./review-hard./test-audit; add --apply only when you want evidence-backed portfolio changes implemented./research-hard./ulw-skip <reason> to pass once./mark-deferred <reason> to bulk-defer all pending advisory findings with a recorded reason — keeps /ulw-report audits accurate./ulw-pause <reason> — credentials/login, external account, hard external blocker, destructive shared-state action awaiting confirmation, unfamiliar in-progress state. Under v1.40.0 no_defer_mode=on (default), taste/policy/credible-approach are NOT pause cases — the agent picks the sane default and ships. Distinct from /ulw-skip (one-shot gate bypass) and /mark-deferred (legacy soft-defer, disabled under ULW execution).Three skills, three different "I can't keep going" cases — NOT interchangeable. The full decision tree (symptoms, when-NOT-to-use, validator-acceptable WHY shapes, v1.40.0 no-defer contract) is the canonical single source of truth in ~/.claude/quality-pack/memory/skills.md → "Deferral-verb decision tree (which one to use)". That file loads into every session, so the model already has it; this page intentionally does not duplicate it.
Quick reference (symptom → verb):
| Symptom | Verb |
|---|---|
| Gate fired but the work is genuinely done (false positive) | /ulw-skip <reason> |
Discovered-scope flagged real findings you're consciously NOT shipping (v1.40.0: refused under ULW execution with default no_defer_mode=on — agent ships inline instead) | /mark-deferred <named-WHY> |
| Blocked on an OPERATIONAL input — credentials/login, hard external blocker, destructive shared-state action, unfamiliar in-progress state | /ulw-pause <reason> |
Escalation order before any of these fires: ship inline → wave-append → reject-as-not-a-defect → pause-for-operational-block. Under ULW with no_defer_mode=on (default), /mark-deferred is unavailable; the agent owns technical judgment and ships.
/ulw-resume — atomically claims the most relevant unclaimed resume_request.json for the current cwd (or matching project_key) and replays the original objective. The SessionStart resume-hint hook surfaces the artifact automatically; /ulw-resume is the explicit claim verb. Run /ulw-resume --peek to inspect first, --list to see all claimable artifacts./memory-audit — classifies entries and proposes rollup moves without moving anything itself./atlas.Note: Specialist agents activate automatically based on your task. You don't need to learn agent names — just describe what you want to accomplish.
See: the "Auto-routed vs. manual escape hatches" subsection in
README.mdfor the full tier breakdown — which specialists auto-suggest under/ulw, which fire from hooks, and which slash commands are escape hatches for standalone use.
What's with the names? The mythology-themed skills (atlas, metis, oracle, prometheus, librarian) carry one-word mnemonics in
docs/glossary.md. Search that page for the verb you want (plan, review, debug, look up docs) to find the matching skill.
Maximum-autonomy professional work mode. Use for non-trivial coding, writing, research, planning, or mixed tasks that should be carried through the right specialist workflow with minimal hand-holding.
Deactivate ultrawork mode for the addressed session. Clears its workflow and enforcement state without affecting concurrent Claude sessions.
Adaptive project evaluation that maps the task's actual risks, assembles the smallest useful specialist team, and fills material coverage gaps before synthesis.
Guided oh-my-claude configuration via multi-choice prompts — first-time setup walkthrough, post-update review, or ad-hoc inspect/change. Auto-detects mode from `~/.claude/oh-my-claude.conf`. Use when the user asks to install, set up, configure, update, walk through settings, change flags, turn the watchdog or bias-defense on/off, switch model tier, or any phrase like "help me install/update/configure oh-my-claude", "show my settings", "what flags do I have on".
Audit whether a project's tests still earn their runtime and maintenance cost. Maps behavior contracts to owners, identifies redundant or obsolete proof, and recommends KEEP, EXTEND, MERGE, REPLACE, DELETE, or ADD with evidence. Use --apply to implement proven portfolio changes.
Render a markdown digest of recent harness activity — sessions, gate fires, top reviewers, classifier misfires, Serendipity catches, and finding/wave outcomes — over a chosen time window. Use to answer "is the harness actually helping me?". Pass `--share` for a privacy-safe shareable card (numbers + distributions only; no prompt text or free-text reasons).