Skills #

Agent Skills discovery and prompt behavior.

Skills #

SKILL.md files are discovered recursively from the original fen roots plus pi/Agent Skills-compatible locations: ${XDG_CONFIG_HOME:-~/.config}/fen/skills, .fen/skills, bundled fen skills materialized under ${XDG_DATA_HOME:-~/.local/share}/fen/skills/bundled, ~/.pi/agent/skills, ~/.agents/skills, project .pi/skills, ancestor .agents/skills, and common Claude/Codex skill roots. User and project skills are discovered before bundled skills, so a skill with the same name shadows the bundled copy. Set FEN_DISABLE_BUNDLED_SKILLS=1 to skip bundled-skill materialization and discovery. Discovery skips dotdirs, node_modules, and paths matched by .gitignore, .ignore, or .fdignore. Explicit paths can be passed via --skill <path>; --skills <dir> remains a compatibility alias.

Frontmatter is minimal YAML. description is required; name is optional and falls back to the skill directory/file name. disable-model-invocation: true skills are discovered but omitted from the system prompt. Discovered skills are listed in an Agent Skills-style XML block with absolute paths; the model uses the existing read tool to load the body on demand.

Use /skills to pick a discovered skill and show its detail panel, or /skills <name> to jump directly to one skill. Use /skills list to emit a text list of discovered skills, their scopes, visibility, and source paths. Use /skills visible, /skills hidden, /skills builtin, /skills user, /skills project, or /skills cli to emit filtered lists. The skills extension also exposes a discovered-skills introspection snapshot for /extensions skills and runtime diagnostics.