| name | spec-and-ticket-decomposition |
| description | Transform rough intent into an authoritative, testable specification and coherent dependency-aware task candidates. Use when clarifying requirements, defining non-goals and acceptance criteria, or preparing feature plans before task contracts. |
Spec and Ticket Decomposition
Turn intent into execution without hiding decisions in prose.
Source discipline
Classify every input as:
- approved fact/decision
- repository or contract evidence
- inference to validate
- unresolved question with an owner
Conflicts and missing authority remain explicit. Never resolve business, legal, security-policy, or data-ownership ambiguity by convenience.
Specification procedure
- State the problem, affected actors, user/system value, measurable goal, and non-goals.
- Describe required behavior, failure/recovery behavior, permissions, boundary conditions, and state/data lifecycle where relevant.
- Identify controlled contracts, owners, producers/consumers, compatibility, migration, and rollback constraints.
- Capture applicable quality attributes: security/privacy, accessibility, reliability, performance, observability, operability, and compliance.
- Convert every requirement and material risk into observable acceptance criteria.
- Separate approved decisions from proposals; list open decisions with consequence and required owner.
- Check traceability: every approved requirement maps to behavior and evidence; everything else is a non-goal, risk, or open decision.
Completion criterion: implementation behavior is unambiguous, QA can decide pass/fail, and no material decision is disguised as an assumption.
Acceptance-criterion test
A criterion is valid when it names preconditions, observable action/event, expected result, and critical negative/recovery behavior. Use Given/When/Then when stateful interactions benefit. Replace subjective words with a project policy, threshold, or explicit observation.
Decomposition procedure
- Find seams by user-facing outcome, controlled contract, ownership, risk, and independent testability.
- Build the artifact dependency graph before choosing execution order.
- Form coherent task bundles: one outcome, one accountable owner type, bounded scope, stable inputs, and independently verifiable completion.
- Separate architecture/contract decisions, implementation, QA, release, and revalidation when authority differs.
- Avoid splits that create shared mutable checkout work or repeated edits to high-conflict files.
- For each candidate, provide goal, scope/out-of-scope, dependencies, consumed/produced contracts, definition of done, and required evidence.
- Group candidates into execution waves and flag which tasks may safely run in parallel.
- Validate coverage: every acceptance criterion and controlled change has an owner; no task exists without a requirement/risk trace.
Completion criterion: the candidate graph covers the specification exactly—no missing requirement and no speculative task.
External accelerators
Use to-spec or to-tickets when available, then apply this skill's source, acceptance, decomposition, and completeness tests. Their output is draft until it passes these gates.
Output
Produce a compact specification, candidate tasks, dependency graph, wave plan, controlled-contract notes, risk register, and open decisions. The workstream agent owns final task contracts and assignment.