| name | agentic-sdd-tasks |
| description | Break an approved SDD plan into small, bounded, independently verifiable implementation tasks. |
| allowed-tools | Read, Write, Edit, Glob, Grep, Bash |
| risk | safe |
| source | local |
Agentic SDD Tasks
Use this skill only after the architecture plan is approved or the user explicitly accepts the risk of proceeding.
Goal
Create specs/<feature>/tasks.md that an implementation agent can execute without reinterpreting the spec.
Required Reading
.specify/memory/constitution.md
specs/<feature>/spec.md
specs/<feature>/plan.md
specs/<feature>/review.md
specs/<feature>/test-plan.md, if present
- relevant ADRs
Task Format
Each task must include:
- task id
- objective
- source artifacts
- allowed files or directories
- forbidden files or directories
- acceptance criteria
- verification command or observable result
- dependencies
- risk level
- approval needed, if any
- status
Task Design Rules
- Keep each task small enough for one focused implementation pass.
- Prefer vertical slices by user story when possible.
- Put tests before implementation when behavior is risky or TDD is requested.
- Mark parallel-safe tasks only when write sets are disjoint.
- Do not create tasks that require unapproved dependencies, migrations, auth/payment edits, secrets, production config, or deployment.
- Include documentation and contract updates as explicit tasks when behavior changes.
Recommended Sections
specs/<feature>/tasks.md should include:
- track summary
- prerequisites
- task list
- dependency order
- parallel-safe groups
- verification matrix
- blocked or approval-required work
Stop Conditions
Stop before writing implementation-ready tasks when:
- architecture review is not approved
- tasks would require guessing file ownership
- verification commands are unknown
- a task touches high-risk areas without approval
- dependencies between tasks are unclear
Output
Return task file path, number of tasks, blocked items, parallel-safe groups, and the first safe implementation task.