Skip to main content

For Maintainers

Gives repository maintainers a practical path for making projects easier for AI agents to understand.

Agent-ready documentation is not about adding every convention. It is about making the important project facts easy to find, verify, and update.

Baseline Checklist​

Start with these files:

  • AGENTS.md for project instructions.
  • .aiignore for secrets, generated output, large files, and noisy dependencies.
  • CONTRIBUTING.md for human contribution rules.
  • README.md for human project overview.

This is enough for many projects.

Add Persistent Context When Needed​

Add memory or context files when repeated agent sessions need the same project knowledge.

Good candidates:

  • MEMORY.md for stable facts and open follow-ups.
  • cline_docs/ or .roo/ when a tool expects Memory Bank folders.
  • Architecture notes when agents repeatedly need system boundaries.

Keep instructions and learned context separate. Instructions tell agents what to do. Memory records what has been learned.

Add Public LLM Entry Points​

If the project has public docs, add:

  • llms.txt for a short routing file.
  • llms-full.txt when a compact full-docs snapshot is useful.

These files help agents retrieve the right docs without crawling the whole site.

Add Capabilities and Evals Last​

Use SKILL.md when the team has a repeatable workflow that agents should load on demand.

Use EVAL.yaml when agent behavior must be measured repeatedly, especially in prompts, support workflows, classification, or code-review tasks.

Maintenance Rules​

  • Keep each file short enough to review.
  • Link to canonical docs instead of copying long specs.
  • Remove stale rules when tools or workflows change.
  • Treat convention files as reviewed project documentation, not generated chat output.