| name | writing-plans |
| description | Use when an approved PRD needs an implementation-authoritative FSD before code is written. |
Writing Plans
Turn an approved PRD into the FSD authority consumed by /sc-work. Announce: "I'm using writing-plans to create the FSD and goal slices."
When to Use
Use after BRD/PRD approval and before implementation. Do not plan from memory, invent missing product policy, or proceed while an OPEN-* blocker prevents a safe technical contract.
Route
- Search before loading: locate the approved PRD, qualified BRD refs, existing FSD/state, accepted ADRs, nearby code/tests, and relevant solution notes.
- Load FSD authoring for evidence selection, risk branches, contract shape, acceptance, and handoff.
- After the contract is stable, load goal slicing only when defining
GOAL-* packets, a task ledger, or a measured wide refactor.
- Read only the needed sections of the full FSD template; never preload it merely to orient.
Invariants
- Save the primary authority to
docs/fsd/fsd-<feature>.md; a companion plan never outranks it.
- Trace qualified BRD/PRD IDs, approved
TDEC-*, optional accepted ADRs, tests, and goals. Missing authority becomes an OPEN-* blocker.
- State exact contracts, affected paths, failure modes, rollback, verification commands, and expected evidence.
- Prefer a thin vertical slice that delivers one independently verifiable behavior.
- A wide refactor is an exception: name the shared seam, enumerate affected callers, bound the file surface, and provide verification covering every caller plus migration/rollback. “Many files” alone is not justification.
- Never defer compatibility, security, privacy, data, or UI evidence until implementation.
- For UI-bearing scope, use the canonical UI readiness reference, make FSD
Section 8 the Screen & Interaction Contract, and keep exact wire shape in a
delegated versioned machine contract.
- Define a contract enabler when executable assets are missing, then exactly one
first real vertical slice before any dependent scale-out goals.
Red Flags
| Thought | Response |
|---|
| "The FSD is obvious" | Write exact contracts and evidence. |
| "One horizontal layer is cleaner" | Use a vertical slice unless the measured wide-refactor exception applies. |
| "Tests can wait" | Map verification beside each risky goal. |
| "We can choose the package later" | Resolve compatibility and supply-chain risk first. |
Integration
Inputs: agentic-delivery, prd-generator, domain/codebase evidence. Risk branches: compatibility-check, threat-modeling, data-privacy, interface-design. Outputs: plan-verification, issue-workflow, executing-plans, and verification-before-completion.