claws-skill-chaininglisted

# Skill Chaining ## What this is When Skill A invokes Skill B via `Skill({ skill: "B" })` mid-execution, B runs with its full frontmatter, body, and decision logic as if the user typed `/<B>` directly. Accumulated context stays live; B continues the current turn, not a fresh one. See [[claws-mcp-cognition-engine]] for the Skill→MCP variant. ## When to invoke - **Explicit redirect** — intent belongs to a different skill (`/claws-do "ship X"` → chain to planning). - **Capability gap** — current skill hits a step owned by a specialist (`/claws-fix` → chain to diagnose layer). - **Follow-up workflow** — one skill's output is the natural input to the next (plan output → goal contract). ## When NOT to invoke - **User info needed** — surface the question directly; chaining defers without resolving it. - **Depth > 3** — A→B→C→D should become a wave army with explicit sub-workers instead. - **Circular dependency** — A→B→A burns context; extract shared logic into a third skill. ## Steps 1. Identify trigger: redirect, capability gap, or follow-up workflow. 2. Confirm target exists in [[claws-skill-graph]]. 3. Invoke `Skill({ skill: "<target-name>" })`. 4. Let target complete and return a result. 5. Resume calling skill with the returned value. Write `Skill({ skill: "X" })` in a code fence or `→ /X` as a bullet — prose-only descriptions produce no graph edge. ## Examples **`/claws-do` → `claws-plan` → [[claws-goal-tracker]]** (follow-up workflow): "Ship X" needs a plan then a