Orchestrate a multi-agent workflow to plan, implement, test, and verify a new feature.
Scanned 9/8/2026
Install to Claude Code
npx -y skills add nobodyonlyc/skills --skill workflow-feature --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Workflow Feature?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/nobodyonlyc-workflow-feature)More formats (shields.io, HTML) on the badges page.
---
name: workflow-feature
description: Orchestrate a multi-agent workflow to plan, implement, test, and verify a new feature.
---
Feature request: $ARGUMENTS
Multi-agent feature development workflow that coordinates analysts, developers, and testers around a single feature (WIP = 1). Each phase has a detailed playbook in [`references/`](references/); see [`examples/`](examples/) for a full worked run.
## Skills this workflow drives
- [core-explain](../core-explain/SKILL.md) — read and understand existing code/patterns (Phase 1).
- [core-prototype](../core-prototype/SKILL.md) — quick 1–2 screen mock UI for the Phase 1 design gate (escalate to [workflow-prototype](../workflow-prototype/SKILL.md) for a full multi-screen flow).
- [dev-db-designer](../dev-db-designer/SKILL.md) — schema & migration design when DB changes are needed (Phase 1 & 2).
- [check-test-gen](../check-test-gen/SKILL.md) — plan and write the test suite (Phase 1 strategy, Phase 2 execution).
- [dev-be-developer](../dev-be-developer/SKILL.md) / [dev-fe-developer](../dev-fe-developer/SKILL.md) / [dev-cli-tool-developer](../dev-cli-tool-developer/SKILL.md) / [dev-batch-developer](../dev-batch-developer/SKILL.md) — implementation, chosen by component type (Phase 2).
- [check-code-review](../check-code-review/SKILL.md) — review the diff each loop iteration (Phase 2).
- [check-qa](../check-qa/SKILL.md) — full verification pass before the checkpoint (Phase 3).
- [check-security-review](../check-security-review/SKILL.md) — mandatory security audit in Phase 3 when the change touches auth, external input, secrets, queries, shell, or crypto.
## Phases
1. **Confirmation, Analysis & Planning** → [references/phase-1-analysis-planning.md](references/phase-1-analysis-planning.md)
Confirm the target feature (WIP=1), run Requirements Analyst + Test Strategist in parallel, approve the plan and (if UI) the design.
2. **Code · Test · Review Loop** → [references/phase-2-build-test-review-loop.md](references/phase-2-build-test-review-loop.md)
Iterate: parallel FE/BE implementation → run tests → code review, looping until clean.
3. **Verification & Checkpoint** → [references/phase-3-verify-checkpoint.md](references/phase-3-verify-checkpoint.md)
Run the QA suite, `./harness verify`, session stop, clean, then STOP.
## Hard gates
- **Autonomy mode**: honor the run's mode (`gated` default / `auto`) per [autonomy-mode](../../resources/autonomy-mode.md). In `auto`, ask-user gates below become logged decisions — except the always-stop list, which always halts.
- **Durable task state**: maintain `.harness/tasks/<id>.md` per the [task-state convention](../../resources/task-state-convention.md) — created at start, its **granular DoD checklist updated and committed at every phase boundary AND before marking `passing`**, with acceptance criteria each ticked and mapped to a test. This holds **in `auto` mode too** — auto suppresses ask-user gates, never state-keeping. A task file left at an early step while the feature is `passing` is a process failure. `evidence.md` must be non-empty before verify (hook-enforced, see [task-state convention](../../resources/task-state-convention.md)).
- **Strict File-Based Communication**: Do NOT pass error logs or code snippets via chat messages. When Reviewers/Testers output files to `.harness/reports/`, send ONLY the file path to the Developer subagent.
- Work on exactly **one** feature (WIP = 1). Confirm the target with the user before any work.
- If the feature has UI, get explicit approval of a visual mockup **before** writing implementation code (see Phase 1).
- **Unit tests are mandatory:** the feature ships with UT covering its logic and edge cases, written and passing inside the Phase 2 Code·Test·Review loop. A failing UT loops back through fix → re-run, never gets silenced. A feature with no UT (or existence-check-only "tests") is not done.
- A feature only moves to `passing` after `./harness verify <id>` succeeds with evidence.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!