- Home
- Skills
- Agents & Orchestration
- project autopilot
project autopilot
A multi-agent orchestrator that plans, codes, reviews, and tests features using Claude Code and other agent CLIs.
Free
Works with the AI Agents you already use
project autopilot
Example session with this skill installed
ap run "Implement a responsive navigation bar with a mobile hamburger menu and ensure it passes all existing layout tests."
- Read your context and instructions
- Compiled the project autopilot
- Generated the document
Mission started: 001-responsive-nav. I have initialized the planning phase and assigned the roles.
Mission Progress: 001-responsive-nav
| Phase | Role | Status | Output/Evidence |
|---|---|---|---|
| Planning | planner | COMPLETE | PLAN.md and TASKS.md generated. |
| Plan Review | plan_reviewer | PASSED | Reviewer approved 3 tasks with no findings. |
| Implementation | implementer | IN_PROGRESS | Task 1: src/components/Nav.tsx created. |
| Code Review | code_reviewer | PENDING | - |
| Gate | system | PENDING | Will run npm test. |
| Verification | verifier | PENDING | Will check Playwright criteria. |
Current Status: The implementer is currently writing the CSS modules for the mobile drawer. The planning phase identified a dependency on framer-motion which was already in package.json.
Next steps
- Monitor implementation progress in
.autopilot/missions/001-responsive-nav/WATCH-REPORT.md. - Once the final review is complete, run
git diffto inspect the changes. - Run
git committo finalize the mission.
Nav.tsx
PDF · document
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Managing complex coding tasks across multiple agent sessions often leads to half-finished features, untested code, or recursive loops that drain your token budget. Project Autopilot provides a structured multi-agent delivery pipeline that forces every request through a rigorous sequence of planning, implementation, and multi-stage review.
What it does
- Multi-Role Orchestration coordinates separate agents for planning, implementation, code review, and verification to ensure no single agent is grading its own work.
- Automated Gatekeeping executes user-defined test commands after implementation and only proceeds if the codebase remains stable.
- Read-Only Containment uses
bubblewrapon Linux/WSL2 to force reviewers and verifiers into a read-only state, preventing accidental or malicious changes during the audit phase. - Stateful Resumption tracks mission progress in
.autopilot/state.json, allowing the system to resume from the exact point of failure or interruption. - Browser Verification integrates with Playwright to verify UI-based tasks in a live browser environment.
How it works
- Initialize the project using
ap setupto configure roles, preferred CLIs (Claude Code, opencode, Codex), and test commands. - Start a Mission by running
ap run "your request"to trigger the planning and implementation pipeline. - Monitor Progress via the
WATCH-REPORT.mdfile which updates in real-time as agents complete tasks and pass reviews. - Human Handoff occurs once the final review is successful, leaving the verified changes in the working tree for a final manual commit.
Frameworks & tools
This tool is built for Node.js environments and integrates with Git. It drives external CLIs including Claude Code, opencode, Codex, and Antigravity. It supports SpecKit for workflow management and Playwright for end-to-end browser testing.
Why this beats prompting it yourself
Manual prompting lacks the state management required for long-running tasks that span hours or multiple sessions. This skill enforces a "defense in depth" review strategy and hardware-level read-only guards that a standard chat interface cannot replicate.
Use cases
- Implement complex features that span multiple files and require architectural planning.
- Run autonomous bug-fix cycles that require verification via existing test suites.
- Perform ad-hoc code reviews or plan audits using specialized, read-only agent roles.
- Execute long-running background coding tasks via detached sessions on Linux/WSL2.
Known limitations
The tool does not support automatic git commits or pushes. Read-only enforcement requires bubblewrap on Linux/WSL2; Windows uses disposable-copy isolation which is not OS-enforced.
Requires Node.js 18+ and Git. No automatic commits. Strong containment requires bubblewrap on Linux. GSD and Cursor integrations are currently unreviewed.
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 any payment.
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
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 20 days ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts