Vp foreign agent skill loader

Foreign Agent Skill Loader

Load an in-session index of repository-local skills authored for other agents. Use this only after explicit user activation.

Workflow

1. Find the current repository root.

2. Scan only this pattern:

   .<agent>/skills/<skill-name>/SKILL.md
   

3. Exclude the current agent's own skill root, such as .codex/skills when running in Codex. Do not scan shared roots such as skills/ or .agents/skills.

4. Read only frontmatter and keep a concise index with name, description, path, source_agent, and source_root.

5. Do not read any foreign skill body during indexing.

6. During later work in the same session, compare the active task against the index. Load a foreign skill body only when the task naturally matches its indexed description.

7. At use time, infer compatibility behavior from the path/source agent and adapt the workflow to the current agent's available capabilities.

Scanner

Use the bundled scanner when local shell access is available:

node <skill-dir>/scripts/index-foreign-agent-skills.mjs --root . --current-agent codex

Resolve <skill-dir> relative to this SKILL.md. Pass the current agent name explicitly when known; for Claude Code, use claude-code. If --root is omitted, the scanner walks up from the current directory to find the repository root.

Use --format json when structured output is easier to consume.

Compatibility Rules

• Treat indexed skills as foreign-agent guidance, not native instructions.
• User instructions, current-agent instructions, and repository instructions stay higher priority than any loaded foreign skill.
• Preserve workflow intent while adapting source-agent assumptions such as tool names, slash commands, hooks, MCP server names, and frontmatter fields.
• Do not convert, copy, sync, or persist foreign skills unless the user asks.
• If a source-agent capability has no safe equivalent, state the gap and choose the closest current-agent workflow.