skill-retrieval.yaml 1.5 KB

12345678910111213141516171819202122232425262728293031
  1. name: skill-retrieval
  2. repo: https://github.com/moonlight-lupin/agent-skills
  3. sha: 79801c69e797f5f8efd619f8f451ca89d18c7240
  4. subdir: plugins/skill-retrieval
  5. description: >-
  6. BM25-based skill retrieval for large skill libraries. An llm_request middleware rewrites
  7. the outgoing request's <available_skills> block to names only (the same bytes every turn,
  8. so prompt caching is unaffected), and a pre_llm_call hook appends the top-K matching
  9. skill descriptions to the current user message. Uses public plugin surfaces only
  10. (register_middleware, register_hook); reads SKILL.md frontmatter from the profile's skill
  11. roots and imports a few agent.prompt_builder / agent.skill_utils helpers read-only.
  12. pyyaml is the only dependency. Disclosure — optional Jev rerank, off by default: with
  13. SKILL_RETRIEVAL_RERANK=jev and TYPESAFE_API_KEY set (read from the environment only), each
  14. turn POSTs the user message (injection patterns stripped, capped at 2000 chars) and up to
  15. 12 shortlisted skill descriptions (400 chars each) to api.typesafe.ai (3 s timeout, no
  16. retries, BM25 order kept on error), and appends a local JSONL log (timestamp, session id,
  17. 200-char query excerpt, both orders) under $HERMES_HOME/data/jev-trial/, rotated past 5 MB.
  18. With the rerank off it makes no network calls and writes no files.
  19. maintainer: moonlight-lupin
  20. tier: community
  21. category: tools
  22. requires_hermes: ">=0.21.1"
  23. version: "0.6.0"
  24. capabilities:
  25. provides_tools: []
  26. provides_hooks:
  27. - pre_llm_call
  28. - on_skill_lifecycle
  29. provides_middleware:
  30. - llm_request
  31. requires_env: []