How to add a custom harness

This guide shows you how to add a harness the five built-ins do not cover. Run the commands from a directory that has skillprism.yaml, or from a directory under it. The definition fields are in the harness reference.

Pick an id that is not claude, codex, opencode, factory, or pi. A file that reuses a built-in id replaces that built-in. Do not use claude-code or droid unless you mean to replace that alias: an exact id wins over the alias.

skillprism init harness my-agent

This writes harnesses/my-agent.yaml. build, add, and update load every .yaml and .yml file in that directory.

Edit project_scope_path and user_scope_path so they match the directories that agent reads. Leave skill_filename as SKILL.md unless the agent expects another name.

Add the id to skillprism.yaml, or a plain build will not include it:

harnesses:
  - claude
  - my-agent

Compile it:

skillprism build -H my-agent

The skill file is dist/my-agent/<skill>/SKILL.md.

To give that harness a JSON index, add the path fields to the existing paths map:

manifest_scope_path: .my-agent
manifest_filename: index.json

Add this map. Do not add a format key:

manifest:
  template: |
    {
      "name": {{ skill_name | tojson }},
      "description": {{ skill_description | tojson }}
    }

Apply tojson to each interpolated string and leave the filter unquoted. build renders one entry per skill, checks that the aggregated file is JSON, and writes dist/my-agent/.my-agent/index.json. A broken template fails the build before any output is written.