AGENTS.md gives coding agents repository-specific instructions. Entelligence can surface its status and include it in the documentation synchronization workflow.
What belongs in AGENTS.md
- Repository purpose and boundaries
- Build, test, lint, and run commands
- Directory or package ownership
- Architecture constraints
- Required review and validation steps
- Security and secret-handling rules
- Generated-file and migration guidance
- Local terminology
Keep instructions executable and current. Avoid product slogans, broad policy that belongs elsewhere, and secrets.
Recommended structure
Review generated instructions
- Verify commands on a clean environment.
- Confirm paths and package names.
- Check that instructions do not weaken security or review policy.
- Remove secrets and internal credentials.
- Keep high-risk writes behind explicit confirmation.
- Commit through the normal pull-request process.
- Assign an owner.
Synchronization
Use the repository’s Configure action to select the target agent-docs folder and schedule weekly or monthly synchronization. Review the AGENTS.md status in the repository table after each material change.
Agent instructions can directly influence automated code changes and shell commands. Treat an AGENTS.md update as code: review the diff, test the commands, and protect the branch.
Avoid conflicts
Multiple instruction files can apply at different directory levels. Keep the most specific rule closest to the code it governs and remove contradictions. When Entelligence-generated content and a manually maintained file differ, resolve ownership rather than allowing repeated overwrites.
Verify
Ask a coding agent to state the applicable instructions before making a harmless test change. Confirm it identifies the correct commands and boundaries.
Next step
Configure sync and export or read troubleshooting.