If you're building an AI agent, IDE, or CLI tool, adding ACS support makes your tool compatible with any project that uses ACS.
The full behavioral specification is in spec/v1/10-behavior.md.
Minimum viable implementation (required):
- Scan for
.agents/main.yamlat session start - Parse the manifest to determine active layers
- Load
context/project.mdinto session context - Enforce
permissions/policy.yamlrules on file operations - Surface skill and command names/descriptions (catalog) to the model
Recommended:
- Implement progressive disclosure (Tier 1/2/3 loading)
- Protect context from compaction
- Handle monorepo nested
.agents/with correct precedence
Once implemented, declare compatibility in your docs:
ACS-compatible (v1.0)
And optionally open a PR to add yourself to community/ADOPTERS.md.
Use the schemas in schemas/v1/ to validate files in your tool's linting or import logic.
See reference-impl/ for parser and validator examples in Python and TypeScript.