spec-anchored-agentic-development: Skill for Claude Code

Ticket readiness review

Review the ticket contracts and, when available, the complete proposed set. Do not review the implementation plan or code. Do not edit the tickets.

Classify every finding as [BLOCKER], [SHOULD], or [NIT]. Include the evidence and a concrete correction.

Use these tests:

Could a fresh worker execute this task without inventing semantics, scope, dependencies, evidence, or permissions?

Is this issue a coherent delivery task, rather than a requirement, component, test, specialist pass, or implementation step promoted into the tracker?

Does the complete set minimize coordination cost without combining independently acceptable outcomes?

Does a broad outcome label hide capabilities that can be reviewed, accepted, recovered, reverted, or verified independently?

Per-ticket checks

1. Effective spec and stable IDs — the ticket points to the effective spec revision and uses complete stable IDs. Missing, unknown, positional, or recreated normative references are [BLOCKER].

2. Problem is a claim — the current problem is stated without presuming an unproven diagnosis, and the worker may establish NO_CHANGE_REQUIRED. A demanded patch regardless of current behavior is [BLOCKER].

3. One primary outcome — assess size by the number of independently demonstrable outcomes, not by files, steps, requirements, or estimated diff size:

   • no independently useful outcome means the task is too small and should normally be merged with the task that completes it;
   • one primary outcome is the intended size;
   • multiple independently acceptable outcomes mean the task is too large and should be split at those boundaries.

   Supporting behaviors, failure cases, evidence, and hardening for the same primary outcome do not count as separate outcomes.

   Apply the umbrella-outcome test: expand broad labels such as safe lifecycle, platform foundation, or evidence system into the concrete capabilities delivered. If a reviewer could accept one capability and reject another without contradicting the stated outcome, the task is too large and must be split at those independent boundaries. A broad noun phrase does not create cohesion.