This project uses a file-based context injection system to define AI personas and modular capabilities.
This follows the specification at https://agents.md.
The AGENTS.md file in the root directory defines the global instructions and persona for the agent.
- Injected Verbatim: The entire content is added to every prompt turn.
- Singleton: Only one
AGENTS.mdis active at a time. - Usage: Define project rules, coding standards, and long-term goals.
This follows the specification at https://agentskills.io/.
Skills are located in the .agents/skills/ directory. Each skill is defined by a directory containing a SKILL.md file.
- On-Demand: Skills must be explicitly activated using
/skill activate <name>. Theheyhotword can be used to activate a skill for a single turn. - Automatic: Skills can also be activated by the LLM using
tools.use_skill. - Structured: Uses YAML frontmatter for metadata (name, description).
- Extensible: Allows adding specialized knowledge (e.g., C++ expert, DBA) without bloating the global context.
| Feature | AGENTS.md | SKILLS.md |
|---|---|---|
| Scope | Global (Root) | Modular (.agents/skills/) |
| Activation | Automatic (if loaded) | Explicit (/skill activate) |
| Format | Verbatim Markdown | Markdown with Frontmatter |
| Purpose | Core Persona | Specialized Expertise |