copilot-cli

Execution, flag assembly, prompt framing, and output parsing live inside the dispatched sub-skill — not here.

Operation Routing Table

Operation	Sub-skill	Use for
review	review/	Code review of a change set; structured findings + raw markdown
ask	ask/	General query or advice; plain text answer
explain	explain/	Explain a code region or file; explanatory markdown

How to Route

1. Parse the task → identify operation from the table above.
2. Operation unclear → ask the caller; in non-interactive flows return Status: NEEDS_CLARIFICATION.
3. Load + dispatch the identified sub-skill; pass the full task and all caller-supplied context.
4. Task spans multiple operations → run the primary operation; report remaining operations to the caller without dispatching them.
5. Return the sub-skill's structured result unchanged to the caller.

Result Envelope

Status: CLEAN | FINDINGS | OK | ERROR | UNAVAILABLE | NEEDS_CLARIFICATION
<sub-skill result fields>
Source: <sub-skill name>

UNAVAILABLE and NEEDS_CLARIFICATION originate from the router. All other statuses pass through from sub-skills unchanged.

Cache Integration

Before dispatching any sub-skill, check the capability cache (see capability-cache/SKILL.md):

• Cache HIT with result: unavailable → skip all CLI invocations; return Status: UNAVAILABLE to the caller immediately.
• Cache HIT with result: available → use cached model list without re-probing.
• Cache MISS → proceed to dispatch; the probe happens inside the sub-skill; cache is populated by the sub-skill on first run.

Rules

• Do NOT execute any copilot command. All execution is inside sub-skills.
• Do NOT inject flags, model names, prompts, or defaults. Sub-skills own those choices.
• Do NOT log or inspect Copilot's raw output. Only the sub-skill's structured result is returned.
• Do NOT attempt installation or auth recovery if copilot is unavailable — surface the sub-skill's error and stop.
• Sub-skill missing → report that and stop; do not improvise.

Related: review, ask, explain, capability-cache