| name | feature-workflow |
| description | Orchestrate a full feature through staged planning and parallel implementation by sequencing the other framework skills. Trigger on "feature workflow", "run the full workflow", "implement this epic end to end", "/feature-workflow", or when a large feature must go through requirements → plans → implementation → integration → review. |
| deps | {"plan_requirements":"../plan-requirements/SKILL.md","plan_be":"../plan-be/SKILL.md","plan_db":"../plan-db/SKILL.md","plan_domain":"../plan-domain/SKILL.md","plan_fe":"../plan-fe/SKILL.md","clone":"../clone/SKILL.md","do_ui":"../do-ui/SKILL.md","hashy":"../hashy/SKILL.md"} |
ROLE
Orchestrator for large features. Drive the work through staged planning and
layered implementation by invoking the other framework skills — never
re-implement what they do. Keep control at each stage; prefer iterative execution
with a human checkpoint over a single expensive one-shot.
INPUTS
Check before starting. Missing required → list all gaps once, then continue.
| # | Input | Required | Rule |
|---|
| 1 | Requirements | Yes | feature.md-style doc or epic text. Missing → run plan-requirements. |
| 2 | Ideal example | Yes | Module/file whose style implementation mirrors. Missing → ask. |
| 3 | API style | Cond. | REST or RPC, needed before plan-be. Missing → ask at that stage. |
| 4 | DB engine | Cond. | Needed before plan-db. Missing → ask at that stage. |
| 5 | Target module | Yes | Where the implementation lands. Missing → ask. |
DEFINITION OF "WORKING"
Track these as the success metrics for the feature:
- Tests pass.
- The project builds.
- Functionality works manually.
- TDD scenarios are satisfied.
- Acceptance criteria are met.
STAGES
Run sequentially unless marked parallel. Confirm at each checkpoint before
spending on the next stage.
- Requirements — ensure a consistent requirements doc exists; if not, run
/plan-requirements (deps.plan_requirements). Output: business goal, scope,
business rules, constraints, acceptance criteria.
- Plan BE — run
/plan-be (deps.plan_be): endpoints/procedures, contracts.
- Plan DB — run
/plan-db (deps.plan_db): schema, migrations, relations.
- Plan Domain — run
/plan-domain (deps.plan_domain): domain models + events.
- Plan FE — run
/plan-fe (deps.plan_fe): views, components, state, API calls.
- Define the seams (facades) — lock the contracts/DTOs/events/interfaces and
the cross-layer communication rules that let layers proceed independently.
- Parallel implementation — once seams are fixed, implement layers against
them, parallel where possible:
- backend logic, DB model, domain, and frontend each to the agreed facade;
- use
/clone (deps.clone) against the ideal example for code style;
- use
/do-ui (deps.do_ui) for pixel work from a graphic.
- Integration + tests — wire layers together; run unit, integration,
contract, and e2e tests.
- Verify + iterate — run tests, check the success metrics, manual test, fix,
repeat until the feature works.
- Stamp docs — if the ideal example or its
AGENTS.md changed, re-run
/hashy (deps.hashy).
RULES
- Orchestrate only — delegate every stage to the owning skill; do not duplicate
their logic here.
- Keep a human checkpoint between stages; favor iterative control over one-shot.
- Fix the facades/contracts before any parallel implementation begins.
- Parallelize only layers that share a locked seam.
- Hold the line on architecture: each layer separated by TS interfaces, swappable
without breaking others.
- A stage is done only when its output is usable by the next stage.
- The feature is done only when all "working" metrics pass.
- If something is unclear, ask grouped clarification questions before proceeding.