What this agent does
As a workspace accumulates skills, "do we already have something for this?" becomes a real question nobody can answer from memory. This agent indexes every skill in your workspace and answers that question instantly — point it at a task and it tells you which skill (if any) already covers it, or confirms there's a gap.
stage-1-index-build— walks every skills directory in your workspace, parses each skill's frontmatter and description, and builds a single searchable index.stage-2-match-query— given a plain-language description of a task, searches the index and returns the best-matching skill(s) with a confidence note, or reports that no existing skill covers it.
The index is rebuilt incrementally, so adding a new skill to your workspace makes it discoverable on the very next query without a full re-scan.
Requirements
- Works on OpenClaw — this agent is built specifically around OpenClaw's project-scoped skills convention (AgentSkills-format frontmatter).
- No external API required — indexing and matching both run on your existing OpenClaw model access.
Setup
- Unzip the package into your workspace, keeping the folder structure intact.
- List the directories where your skills live (defaults to the standard OpenClaw skills locations).
- Run the starter prompt to build the initial index.
- From then on, just ask "which skill handles X?" and the agent queries the index directly — no need to re-run indexing manually unless you add skills outside the watched directories.
What you get
- A single searchable index of every skill across your entire workspace
- An instant answer to "do we already have something for this?" instead of relying on memory
- A confidence-scored match (or an honest "no existing skill covers this") for any plain-language task description
- An index that stays current incrementally as you add new skills, with no full re-scan required