feat: add search_mixedbread built-in skill (Mixedbread Basic Web Search)

[samsja]

Summary

Adds a search_mixedbread built-in skill to nano-rlm (alongside edit and search), enabled per run via RLM_SKILLS and pre-imported into the IPython kernel as await search_mixedbread(query=..., num_results=5).

It performs Mixedbread Basic Web Search — POST https://api.mixedbread.com/v1/stores/search with store_identifiers: ["mixedbread/web"] — and returns each result as title, URL, relevance score, and the page-content chunk text (a large chunk of the page, often the whole page for short pages).

Changes

• src/rlm/skills/search_mixedbread.py — new skill module (search, run, format_results), needs MIXEDBREAD_API_KEY
• src/rlm/skills/__init__.py — registered in _BUILTIN_SKILLS
• src/rlm/prompt.py — SEARCH_MIXEDBREAD_SKILL_PROMPT addendum when enabled
• README.md — docs: intro, env-var table (MIXEDBREAD_API_KEY), Skills section
• tests/test_builtin_skills.py — 5 new tests (enable stub, missing-key error, formatting)

Enable

RLM_SKILLS=search_mixedbread   # + MIXEDBREAD_API_KEY in the env

Verified

• 13/13 test_builtin_skills.py tests pass (the 4 full-suite failures are pre-existing on a clean tree — they require the example say skill CLI)
• Live-tested against the real API: returns full page-content chunks (~5.6K chars for a Wikipedia page)

---

Note

^{Cursor Bugbot is generating a summary for commit b63db81. Configure here.}