| name | skill-orchestrator |
| description | Orchestrate skill-based work by deciding whether to reuse, combine, extend, or create skills for a task. Use when a request spans multiple skills, when you are planning a new skill or workflow, when existing skills may already cover parts of the work, or when a project needs a repeatable skill stack instead of ad hoc prompting. Prefer this skill before creating new skills from scratch. |
Skill Orchestrator
Coordinate skill-based execution as a system, not as isolated one-off triggers.
Core rule
Follow this decision order:
- Reuse an existing skill.
- Combine multiple existing skills.
- Extend an existing skill or wrapper workflow.
- Create a new skill only if a real gap remains.
Do not skip directly to new-skill creation unless earlier options have been checked.
Main workflow
Step 1 — Break the task down
Identify the task's subproblems, outputs, tools, and recurring parts.
Step 2 — Search existing skills
Check local skills first. If appropriate, also check external sources such as ClawHub.
Step 3 — Evaluate coverage
For each subproblem, decide whether coverage is:
Step 4 — Choose the orchestration path
Pick one:
- Reuse: one existing skill covers the work.
- Combine: multiple existing skills cover the work together.
- Extend: an existing skill is close, but needs added references, rules, or wrapper logic.
- Create: no acceptable coverage exists; build a new skill.
Step 5 — Produce a clear execution plan
State:
- chosen path
- skills involved
- role of each skill
- gaps or risks
- final artifact to create, if any
Skill composition model
Treat multi-skill work as explicit composition.
Relationship types
- Prerequisite: a skill that should run first
- Companion: a skill that strengthens another skill during the same task
- Escalation: a stronger or deeper skill used when the first path is insufficient
- Builder: a skill used to create or improve another skill
Skill stack guidance
For complex recurring work, define a lightweight skill stack with ordered roles.
Example patterns:
- discovery → install/evaluate → build
- governance → workspace cleanup → skill refinement
- research → synthesis → automation
Keep the stack explicit. Do not rely on vague intuition about which skill should come next.
Conversational skill generation
When a new skill may be needed, do not jump straight into writing SKILL.md.
Use this sequence:
- decompose the task
- inspect existing skills
- identify the true gap
- decide whether to extend or create
- create the smallest sufficient skill
- test on a real use case
- refine from actual failure or friction
Hard constraints
- Prefer reuse over reinvention.
- Prefer composition over proliferation.
- Prefer extending an existing good skill over cloning it.
- Create a new skill only when the gap is stable, recurring, and not better solved by project files or workflow templates.
- Keep routing logic explicit and auditable.
Outputs
When orchestrating skill work, produce a concise decision with:
- recommended path: reuse / combine / extend / create
- skills selected
- why this path is best
- missing pieces
- next implementation action
References
Read these references as needed:
references/skill-stack-patterns.md for general stack design and routing patterns
references/real-skill-stack-examples.md for stack examples that map onto existing skills
references/extend-vs-create-and-conflicts.md for extend-vs-create decisions and multi-skill conflict handling
references/fallback-and-degradation.md for stack fallback rules when preferred skills are unavailable, too risky, or too heavy