v2.4.0 - Agent-Optimized AGENTS.md Format
Agent-Optimized AGENTS.md Format v2
This release introduces a significantly enhanced AGENTS.md format designed specifically for AI coding agent efficiency, based on research and collaborative AI design.
Philosophy Change
AGENTS.md is FOR AGENTS, not humans. Human readability is a side effect, not a goal. Every section, format choice, and word exists to maximize AI coding agent efficiency.
New Sections
| Section | Token Savings | Purpose |
|---|---|---|
| Commands (verified ✓) | 500+ tokens | Time estimates, verified working |
| File Map | 3-5 search cycles | Dense dir/ → purpose format |
| Golden Samples | Full rewrites | Reference files with key patterns |
| Utilities List | Duplicate code | "Don't reinvent" table |
| Heuristics | Decision loops | When/Do lookup table |
| Codebase State | Breaking changes | Migrations, tech debt, known issues |
| Terminology | Misunderstandings | Domain-specific terms |
New Features
scripts/verify-commands.sh- Validates documented commands work, prevents "command rot"- "Last verified" timestamp - Enables agent trust scoring
- Language guidance - English default, match code naming to prevent hallucinations
Changes
- Relaxed checkpoint severity (spec has no required fields)
- Added new checkpoints AG-13 through AG-18
- Updated README to distinguish agents.md vs Agent Skills standards
Research Basis
- arXiv: Impact of AGENTS.md on AI Coding Agents - 16.58% token reduction, 28.64% time reduction
- GitHub Blog: Best practices from 2,500+ repos
- Claude + Gemini collaborative design (3 rounds)