Skip to content

AGENTS.md

Latest commit

 

History

History
161 lines (94 loc) · 1.78 KB

AGENTS.md

File metadata and controls

161 lines (94 loc) · 1.78 KB
Raw

AGENTS.md

This repository supports multiple AI coding agents.

Examples:

  • Claude Code
  • Cursor
  • Codex
  • Gemini
  • OpenAI Agents
  • future AI tools

All agents should follow the same workflow.

Source of Truth

Actual source code is always the source of truth.

Documentation must match code.

When documentation conflicts with code:

verify first

then update documentation.

Workflow

Read first:

AGILE_V_PROTOCOL.md

Then:

CLAUDE.md

Then:

.agile-v/STATE.md

Resume from the latest checkpoint.

Before Coding

Always:

  • inspect architecture
  • inspect affected files
  • inspect related requirements
  • identify dependencies
  • identify risks
  • produce implementation plan

Wait for approval.

During Coding

Implement only approved scope.

Prefer:

  • reusable components
  • reusable hooks
  • reusable libraries
  • reusable utilities
  • reusable types

Avoid:

  • duplicate logic
  • duplicate API calls
  • duplicate state
  • duplicate schemas

Rendering

Prefer SSR.

Keep layouts server-rendered.

Only interactive parts should be client components.

Avoid unnecessary loading pages.

Preserve layout stability.

CRUD

Every successful mutation must:

  • persist data
  • invalidate affected caches
  • update optimistic state
  • synchronize UI

Avoid stale data after:

  • navigation
  • back button
  • refresh
  • detail pages

Documentation

Update only affected files.

Examples:

.agile-v/STATE.md

REQUIREMENTS.md

DECISION_LOG.md

VALIDATION_SUMMARY.md

TASKS.md

Do not rewrite unrelated documentation.

Validation

Run relevant validation.

Never claim success without verification.

Record results.

End of Session

Update:

STATE.md

Write:

  • completed work
  • remaining work
  • blockers
  • next exact task

Future agents should resume without reading chat history.