Skip to content

PRD.md

Latest commit

 

History

History
151 lines (109 loc) · 2.95 KB

PRD.md

File metadata and controls

151 lines (109 loc) · 2.95 KB
Raw

PRD: Autonomous Development Supervisor Agent

1. Product Name

Autonomous Development Supervisor Agent

2. Goal

Build an AI-powered supervisor system that controls and validates the full development cycle performed by a coding agent.

The system must not only generate code. It must:

  • Give approvals when a coding tool asks for permission
  • Monitor coding progress
  • Run the project
  • Test the application
  • Detect bugs
  • Create reports
  • Send bug reports back to the coding agent
  • Repeat the cycle until the project is production-ready

3. Main Users

  • Solo developers
  • AI coding tool users
  • Startup builders
  • QA automation teams
  • Agencies building client projects

4. Core Problem

AI coding tools can write code, but users still need to:

  • Approve file edits
  • Approve terminal commands
  • Run the app manually
  • Test buttons and pages
  • Find bugs
  • Take screenshots
  • Tell the coding agent what to fix
  • Repeat the process many times

This wastes time and breaks automation.

5. Solution

Create a Supervisor Agent that works like a developer + tester + reviewer + permission operator.

6. Core Features

6.1 Requirement Manager

The agent must:

  • Read requirements
  • Create task breakdown
  • Create completion checklist
  • Track implemented vs missing features

6.2 Coding Agent Monitor

The agent must:

  • Monitor file changes
  • Monitor terminal commands
  • Track build status
  • Track coding agent progress
  • Detect when the coding agent says the work is done

6.3 Permission Handler

The agent must:

  • Detect permission prompts on screen
  • Locate buttons such as Allow, Approve, Continue, Grant Access
  • Click approval buttons like a human
  • Log all approvals

6.4 Build Validator

The agent must:

  • Install dependencies
  • Run build commands
  • Detect compile errors
  • Capture stack traces
  • Generate build failure reports

6.5 QA Testing Agent

The agent must:

  • Launch the local application
  • Open browser
  • Visit pages
  • Click buttons
  • Submit forms
  • Test navigation
  • Check API calls
  • Check browser console errors
  • Check network failures
  • Capture screenshots

6.6 Bug Reporter

The agent must create structured bug reports with:

  • Page
  • Issue title
  • Steps to reproduce
  • Expected result
  • Actual result
  • Console error
  • Network error
  • Screenshot path
  • Severity
  • Suggested fix area

6.7 Feedback Loop Controller

The system must:

  • Send bug reports to coding agent
  • Wait for code fixes
  • Re-run build
  • Re-run tests
  • Continue until all critical bugs are fixed

7. Final Output

The system must generate a final completion report with:

  • Requirement coverage
  • Build status
  • Test results
  • Bug history
  • Fixed issues
  • Remaining issues
  • Final readiness score

8. Success Criteria

The project is considered done only when:

  • All requirements are marked completed
  • Build succeeds
  • App launches correctly
  • Critical UI flows pass
  • No critical console errors
  • No critical network errors
  • Final readiness score is acceptable