Retrieval skill for searching indexed codebases and documents. Use when you need to find code, understand a symbol or function, search for a concept across a project, or store a summary for future retrieval. Triggers this with 'search the codebase", 'find where X is defined', 'what does X do', 'look up', 'find references to', 'store this summary'.
cercled at 127.0.0.1:7770. Scripts output markdown (formatted by rlm-format). All scoped to CERCLE_SOURCE (defaults to $PWD).
rlm-code-structure "fragment" [n] — Go/Py/JS functions, methods, types — NOT constants. Exact name ranks first. Preview: 20 lines.rlm-search-semantic "concept" 2 0.5 — Always pass limit=2 min_similarity=0.5. Returns 800-char snippet. Scores are not quality rankings — read the snippet."token" [n]rlm-read-symbol <path> <symbol> — Full body of a symbol from a prior structural result.rlm-context <path> <line> [n] — n lines around a line number from any result. Use for constants and file-scope content.rlm-files ["*/glob"] [n] — File list. Use when orienting in an unfamiliar codebase.rlm-list-summaries [n] — Prior agent findings. Use at session start when context is cold.rlm-submit-summary "tags" "text" — Persist findings. Always do after non-trivial research.rlm-embed — Wake embed worker after fresh indexing.rlm-delete-summary <id> — Remove a stale summary.rlm-index <path> [source] — Index a directory and wait for embedding.rlm-delete-source [source] — Remove all documents for a source namespace. Use before re-indexing heavily refactored projects.limit=2 min_similarity=0.5.task run../rlm/scripts/download-vectors, restart daemon. Use lexical meanwhile.For endpoint schemas: REFERENCE.md.