orchestrate

Orchestrate

The conductor. Detects where you are in the superpowered workflow, suggests the next phase, ensures artifacts flow between plugins. This skill doesn't do the work — it routes to the right superpower.

Boundaries

This skill MAY: scan for artifacts (read-only), present status, suggest next steps, route to other skills. This skill MAY NOT: edit code, write documents, execute plans, perform reviews. Routing only.

This skill routes. Other skills do.

The Full Loop

/brainstorm (Perception)
    ↓ writes: docs/brainstorms/YYYY-MM-DD-{topic}-brainstorm.md
/plan (Intelligence)
    ↓ reads brainstorm, writes: docs/plans/YYYY-MM-DD-{topic}-plan.md
/work (Creation)
    ↓ reads plan, executes tasks, commits, ships
/review (Intelligence)
    ↓ reads changes, produces findings
/compound (Knowledge)
    ↓ captures: docs/solutions/{domain}/{topic}.md

Not every task needs the full loop. Match flow to task size.

---

Phase 0: Keyword Routing

Entry: User invoked /orchestrate or used a keyword.

If the user's intent is clear, route immediately:

User Says	Route To
"I have an idea" / "what should we build" / "explore"	/brainstorm
"plan this" / "how should we build" / "break this down"	/plan
"let's build" / "start working" / "execute" / "implement"	/work {plan-path}
"review this" / "check the code" / "is this ready"	/review
"document this" / "save this fix" / "capture"	/compound
"what's next" / "where are we" / "status"	Continue to Phase 1

If intent matches a route: Suggest the skill directly. Don't scan artifacts first.

Exit: Either routed to a skill, or continuing to artifact scan.

---

Phase 1: Detect Current State

Entry: User asked "what's next?" or intent is unclear.

Scan for workflow artifacts:

# Recent brainstorms (last 14 days)
ls -la docs/brainstorms/*.md 2>/dev/null | tail -5

# Active plans
grep -rl "status: approved\|status: in_progress" docs/plans/*.md 2>/dev/null

# Current branch and changes
git status --short
git log --oneline -5

# Recent solutions
ls -la docs/solutions/**/*.md 2>/dev/null | tail -5

Check the project's CLAUDE.md for a "Toolkit Output Paths" table and use override paths if present.

Memory-Informed Routing

When artifacts are found, surface them with direct paths so the user can jump straight in:

• If a brainstorm exists but no plan: >> Ready for planning: /plan docs/brainstorms/YYYY-MM-DD-topic-brainstorm.md
• If a plan exists but work hasn't started: >> Ready for execution: /work docs/plans/YYYY-MM-DD-type-topic-plan.md
• If recent solutions exist for the topic: >> Related prior art: docs/solutions/{domain}/{topic}.md

This reduces friction — the user sees the exact command to run, not just a phase suggestion.

See ../knowledge/autonomy-modes.md for routing confidence patterns.

Exit: Artifacts cataloged.

---

Phase 2: Assess Phase

Entry: Artifacts scanned.

Signals	Phase	Suggest
No brainstorm, no plan, vague idea	Discovery	/brainstorm
Recent brainstorm, no plan	Planning	/plan {brainstorm-path}
Active plan with unchecked tasks	Execution	/work {plan-path}
Active plan, all tasks checked	Review	/review
Review complete, tests passing	Ship	Push and finalize
Just shipped or fixed something non-trivial	Compounding	/compound
Everything shipped	Done	No action needed

Exit: Phase identified.

---

Phase 3: Present Status

Entry: Phase assessed.

## Workflow Status

**Current phase:** {phase}
**Active artifacts:**
- Brainstorm: {path or "none"}
- Plan: {path or "none"}
- Branch: {branch name}
- Uncommitted changes: {yes/no}

**Recommended next step:** {command with path}
**Why:** {one sentence}

Use AskUserQuestion with:

• question: "Here's where things stand. What would you like to do?"
• header: "Next step"
• options:
  1. label: "{Recommended action} (Recommended)", description: "{why this is the logical next step}"
  2. label: "Something else", description: "Tell me what you'd like to do instead"
  3. label: "Start fresh", description: "Begin a new workflow cycle"
• multiSelect: false

If user selects the recommendation: Guide them to invoke the right skill with the right arguments.

If user selects "Something else": Use AskUserQuestion (header: "Action", question: "What would you like to do?") with the common skill options as choices, then route.

Exit: User routed to the right skill.

---

Partial Flows

Task Size	Recommended Flow
Trivial (typo, config)	Just fix it. No workflow needed.
Small (< 1 hour, clear scope)	/work → commit
Medium (1-3 days, clear approach)	/plan → /work → /review
Large (multi-day, unclear approach)	/brainstorm → /plan → /work → /review → /compound
Exploration (no clear goal)	/brainstorm only

Cross-Plugin Artifact Conventions

How plugins find each other's artifacts:

Artifact	Default Location	Key Frontmatter
Brainstorm	docs/brainstorms/YYYY-MM-DD-{topic}-brainstorm.md	type: brainstorm
Plan	docs/plans/YYYY-MM-DD-{type}-{topic}-plan.md	type: plan, status:, brainstorm:
Solution	docs/solutions/{domain}/{topic}.md	type: solution, domain:, symptoms:

The related: field links artifacts across phases. A plan's brainstorm: points to its brainstorm. A solution's related: points to the plan.

Projects can override these paths — see CONVENTIONS.md "Output Artifacts" section.

What Makes This Superpowered

• Workflow Orchestration (5.2): Routing work through the right tools at the right time.
• Augmentation Vision (5.1): Status view shows human + AI collaboration across the lifecycle.
• Process Decomposition (5.3): Breaking projects into plugin-appropriate phases.
• The compound effect: When the full loop completes, knowledge captured makes the next loop faster.

Anti-patterns

• Orchestrating trivial tasks. Don't run the full loop for a typo.
• Auto-advancing. Never auto-advance between phases. The human decides.
• Forcing the full loop. Partial flows are fine.