- Home
- Skills
- Agents & Orchestration
- Plan, Build, Test, Review: 4-Agent Feature Pipeline
Works with the AI Agents you already use
Plan, Build, Test, Review: 4-Agent Feature Pipeline
Four Claude Code agents plan, build, test and review a feature from a rough idea, with a human gate on the spec and a halt after two rejections.
$9
Plan, Build, Test, Review: 4-Agent Feature Pipeline
Example session with this skill installed
Run the dev-team pipeline. Build a 4-page static marketing site for a florist from the attached brief: home, gallery, events, contact. Astro, Tailwind v4, WhatsApp order link with the real phone number, Romanian copy. Show me the spec before building.
- Read your context and instructions
- Compiled the plan build test
Summary
Excerpt from a real run; business details redacted. Four agents, one rejection loop avoided by routing a defect back before the gate.
Plan. dev-planner wrote pipeline/spec.md: 4 pages, 8 named registry components, 5 design tokens, WhatsApp link spec, OUT OF SCOPE: deployment, backend, real photos. ASSUMPTIONS:
4. Shown at the gate; approved.
Build. dev-coder wrote pipeline/ownership.md (7 workers, exclusive file lists) and fanned out. No questions.md; 4 assumptions logged in build-notes.md. Component swaps: 0.
Test. dev-tester ran 6 checkers and wrote pipeline/test-report.md with pasted evidence: build output, 4 routes, the decoded WhatsApp URL, responsive checks at 375/768/1440, token fidelity, and a per-page visual grade:
| Page | Grade |
|---|---|
| index | PASS |
| galerie | PASS |
| evenimente | PASS |
| contact | SLOP, then PASS: a phone button overlapped the social row at both widths; routed to the owning worker and re-graded with overlap = 0 |
Review. dev-reviewer, read-only, re-ran the marker counts and identity grep itself rather than trusting the report. Verdict in pipeline/review-verdict.md:
4 pages build and render, 4 routes, sitemap emitted, 0 original images leaked into dist. 8 registry components installed for real, every usage mounted client:visible. Tokens carried verbatim. WhatsApp link decoded to the exact template text with diacritics intact. MATCHES THE PLAN: YES, two minor non-blocking accuracy defects in the punch-list.
Artifacts left behind:
pipeline/spec.md
pipeline/ownership.md
pipeline/build-notes.md
pipeline/test-report.md
pipeline/review-verdict.md
Next steps the run reported
- Replace 6 placeholder testimonials marked
// FAKE-REPLACE - Confirm 3
[CONFIRM PRICE]markers with the owner - Deploy, out of scope by design
Connects securely to your tools. The creator never sees your data.
About this skill
What you get. A feature or small app built from your idea, plus a pipeline/ folder of artifacts: spec.md, build-notes.md, test-report.md, review-verdict.md (and failure-report.md if the run halts).
How it runs. A planner (opus) writes an exact spec with GOAL, FILES, BUILD STEPS, EDGE CASES, OUT OF SCOPE, ASSUMPTIONS and DONE WHEN. You approve or edit it at a mandatory gate. A coder (sonnet) builds exactly that, a tester (sonnet) writes and runs adversarial tests, and a reviewer (opus) issues APPROVED or REJECTED with numbered reasons. Rejections go back to the coder at most twice, then the run halts with a failure report instead of looping. For hands-off runs, embed an approved spec in the prompt and declare unattended mode; every assumption is then logged in build-notes.md. Works for any stack. If a tier 1, 2 or 3 website skill is installed, the planner applies it.
What's in the package.
SKILL.md(42 lines): orchestrator, flow, spec formatagents/: 4 agent files (dev-planner, dev-coder, dev-tester, dev-reviewer) plus install notes- README with an example prompt; CHANGELOG
Honest limits. The human gate is mandatory unless you declare unattended mode. Output quality tracks the quality of your idea; vague ideas produce long assumption lists. No deployment and no actions outside the project directory. Needs Claude Code and opus and sonnet access (or edit the model: lines).
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Requires dev-planner, dev-coder, dev-tester, and dev-reviewer agents. Maximum 2 rejection loops before halting. Artifacts must be stored in a project-root /pipeline/ folder.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 23 days ago
- Software
- shadcn
Node.jsreact-doctor