Basemind Tools

When basemind is available (its MCP tools or the basemind CLI), prefer it over grep, file reads, and naked git for structural and historical questions. Its tools return paths, line numbers, and signatures — a fraction of the tokens of reading source. Fall back to shell/grep/git only when basemind cannot answer.

• Outline a file before opening it, then read only the span you need (outline; add l2: true for calls and docstrings).
• Find a definition with search_symbols instead of grep.
• Find call sites with find_references (any name) or find_callers (a specific definition) instead of grepping; map structure with call_graph, dependents, find_implementations.
• Search the tree with workspace_grep instead of shelling out to ripgrep.
• Use the git tools — recent_changes, commits_touching, symbol_history, blame_file / blame_symbol, diff_file / diff_outline, hot_files — instead of git log / git blame.
• Use search_documents and the document pipeline for RAG, extraction, and NER over PDFs/docs; use web_scrape / web_crawl / web_map for web content.
• After making edits, run rescan to refresh the index instead of reconnecting. Do not re-read a file basemind already mapped.
• When collaborating with other agents in the repo, use the comms tools (room_list, room_join, room_post, room_history, inbox_read, message_get) to coordinate.
• Any CLI flag carrying prose with code identifiers — a --body, a commit message, a comment posted through a tool — comes from a file or a quoted heredoc, not an inline double-quoted string: `, $(...), ${...}, and ! are all substituted by the shell inside double quotes, silently deleting or mangling every backticked identifier in the text.