Skills Reference

All skills with category grouping and $name usage syntax.

Canonical Workflow

$autopilot is the first-class canonical orchestrator. Its defining default chain is $deep-interview -> $ralplan -> $ultragoal. Each stage is also independently invocable when its input contract is already satisfied.

1. $deep-interview - clarify intent, boundaries, and non-goals when the task is still fuzzy.
2. $ralplan - approve the plan and tradeoffs before execution; completion/skipping requires recorded Architect review evidence followed by Critic review evidence, not just PRD/test-spec files.
3. $ultragoal - execute and verify through durable goal/ledger checkpoints; launch $team only when parallel lanes are warranted.
4. $code-review then $ultraqa - finish with merge-readiness and adversarial QA gates.

Execution Modes

• $autopilot - supervised end-to-end orchestration through $deep-interview -> $ralplan -> $ultragoal
• $team - coordinated multi-agent execution for the approved plan
• $ultragoal - durable multi-goal planning and execution over .omx/ultragoal artifacts plus Codex goal handoff
• $visual-ralph - measured visual-reference loop; Removed in OMX 0.21: $visual-verdict
• $ultraqa - test/verify/fix loop
• Removed in OMX 0.21: $ralph, $ultrawork, $pipeline (sunset stubs for one release)

Planning & Clarification

• $deep-interview - requirements clarification before planning or execution
• $ralplan - planner/architect/critic consensus planning with durable Architect→Critic approval evidence before execution handoff
• $plan - supporting strategic planning when you are not using the canonical path
• $best-practice-research - bounded official/upstream best-practice evidence wrapper when current external guidance affects correctness
• $deep-interview is the independent, resumable Socratic requirements stage before $ralplan; it is not an alias for $plan --interview.
• Removed in OMX 0.21: $prometheus-strict (sunset stub; use $plan).

Research and Planning Boundaries

• $best-practice-research is the ordinary pre-planning wrapper for external official docs, upstream references, standards, release notes, and best-practice checks. It produces cited evidence for a caller; it is not the architecture plan.
• $autoresearch is for a bounded validator-gated research deliverable or artifact. Use it when the research output itself must pass an explicit evaluator before the loop can finish.
• Removed in OMX 0.21: $autoresearch-goal (sunset stub; use $autoresearch)
• $plan is where architecture and implementation planning are synthesized from requirements and prior evidence. Autoresearch findings gathered before planning should feed into $plan as evidence, not appear as a final architecture component unless the user explicitly asks for ongoing research automation.

Agent Shortcuts

• $analyze - read-only deep analysis with ranked synthesis
• $code-review - merge-readiness review
• $design - maintain repo-local DESIGN.md
• $visual-ralph - measured visual-reference implementation loop
• $git-master - alias for the git workflow surface
• Removed in OMX 0.21: $deepsearch (use $analyze), $tdd, $build-fix, $security-review and $review (use $code-review)

Utilities

• $cancel, $doctor
• $skill
• $hud, $omx-setup, $configure-notifications
• $worker (internal) - team worker protocol / lifecycle
• $ai-slop-cleaner - Run an anti-slop cleanup/refactor/deslop workflow
• $ask - Ask a local external advisor CLI (Claude or Gemini) and capture a reusable artifact
• $performance-goal - Run an evaluator-gated performance optimization workflow over Codex goal mode with durable OMX artifacts and safe goal handoffs.
• $wiki - Persistent markdown project wiki stored under repository omx_wiki with keyword search and lifecycle capture
• Removed in OMX 0.21: $help (use $omx-setup), $note, $trace, $ralph-init (use $ultragoal)

Usage Examples

$deep-interview "clarify the auth change"
$ralplan "approve the auth plan and review tradeoffs"
$ultragoal "carry the approved plan to completion"
$team 3:executor "execute the active Ultragoal story in parallel"
$best-practice-research "find current official best practices for this API"
$code-review "review current branch"
$doctor

Visual Verdict Guidance

• Use for visual tasks with reference image(s) + generated screenshot.
• Pass threshold target: 90+.
• Use pixel diff / pixelmatch overlays only as secondary debugging support.