Skills (SKILL.md)¶

This repository supports generating Agent Skills (SKILL.md) into downstream projects.

Where skills live¶

agentsmgr populate project ./defaults writes skills under the project’s configuration directories:

.auxiliary/configuration/coders/<coder>/skills/<skill-name>/SKILL.md

These directories are exposed via the existing coder symlinks at project root, so tools discover skills in their expected locations:

.claude/skills/<skill-name>/SKILL.md
.codex/skills/<skill-name>/SKILL.md
.gemini/skills/<skill-name>/SKILL.md
.opencode/skills/<skill-name>/SKILL.md

Notes:

Tools use filesystem discovery; we do not require a separate “skills registry” file.
Generated directories (agents/, commands/, skills/) are ignored by consolidated coder-root .gitignore files, so they remain untracked.

The base SKILL.md format is standardized (YAML frontmatter + Markdown body), but some fields and behaviors are tool-specific.

allowed-tools is mapped and emitted where supported; for other coders it is omitted.
Argument interpolation is not part of the base standard. If a tool supports invocation arguments (e.g., Claude’s $ARGUMENTS), treat it as a tool extension and avoid relying on it for cross-coder skills.

Keep SKILL.md small and focused:

Put “what to do” in the body.
Push lengthy background, examples, or reference material into adjacent references/ files when needed.
Prefer explicit “Inputs” sections that tell the agent what it must ask for if not already present in conversation context.