The body is dense and operational — commands, file paths, inputs/outputs, and 'When to skip' notes — without explaining concepts Claude already knows (no 'what an SDK is' or 'how ts-morph works' padding). It is above a 2 because nearly every token is actionable context, and there is no verbosity beyond a minor '// turbo' annotation in code blocks.

Every deterministic stage gives an executable command ('npm run capture', 'npm run generate', etc.) with concrete outputs and file paths, and the agent stages point to a specific sibling skill ('stitch-sdk-domain-design'). It is above a 2 because guidance is copy-paste ready rather than pseudocode or abstract direction.

A clearly numbered 9-stage sequence with explicit validation checkpoints (Stage 5/6/7 tests, Stage 8 lock validation) and a feedback loop ('If this fails with a projection error, go back to Stage 2 and fix domain-map.json'), plus an ordering callout for Stage 8. It is above a 2 because validation and error-recovery loops are explicit, not implicit.

Sections are well organized and the agent-only Stage 2 is cleanly offloaded to a one-level-deep sibling skill ('stitch-sdk-domain-design'), but no bundle files exist and the 207-line body is largely monolithic — the Key files table and Stage 9 skill-audit checklist are inline reference material that could live in separate files. It is above a 1 because navigation/structure are clear and the one external reference is well signaled, but below a 3 due to the absence of file-based disclosure for the long inline reference sections.

Names the domain ('Stitch SDK generation pipeline') and actions ('Run the full...pipeline', 'regenerated end-to-end'), but describes one composite action rather than listing multiple specific concrete actions like capture, generate, build, test. It is above a 1 because it is not vague, but below a 3 because the capability is stated as a single abstract pipeline rather than enumerated concrete operations.

Explicitly answers both what ('Run the full Stitch SDK generation pipeline') and when via an explicit 'Use when...' clause with concrete triggers, matching the anchor for clearly answering both what AND when.

Provides relevant natural triggers ('Use when a new tool is added, or the SDK needs to be regenerated end-to-end'), but covers only two trigger phrasings and misses common variations a user might say (e.g., 'add a tool to the MCP server', 'regenerate the SDK', 'update tool definitions'). It is above a 1 because the triggers are natural, but below a 3 due to limited coverage.

Targets a clear niche (the Stitch SDK generation pipeline) with distinct, specific triggers unlikely to fire for sibling skills like usage, readme, or domain-design; it is above a 2 because the trigger 'SDK needs to be regenerated end-to-end' is unambiguous and non-overlapping.