researching-rigorouslylisted

# Skill: researching-rigorously ## Purpose Answer a load-bearing question (or de-risk a decision) by triangulating across the sources actually available, instead of trusting training data or a single tool. The output is a cited finding, not a guess. - **When to use:** any non-trivial fact-check, "is X true in this system?", library/API behavior, locating an object/page/code, parity/source-of-truth questions, validating prior findings. - **When NOT to use:** a value you can read directly from one known file in one step (just read it). ## The source ladder (cheapest-first) Stop as soon as a source settles the question; escalate only if it doesn't. Always cite what you used. 1. **Local sources and repos (offline, fastest)** - Local file clones and mirrors: `rg`/`fd` them FIRST. Directory trees often mirror external hierarchies, so they also tell you where new content belongs. - Read file contents from local clones directly (not via remote API MCPs when a local copy exists). - Any locally available knowledge graphs or pre-built indexes — orient before grepping. 2. **Available knowledge tools** - Graph/navigation tools if present (query/path/explain operations): orient before grepping; verify load-bearing facts against source. - Semantic search indexes, embeddings, or local catalog tools if configured. 3. **MCP servers for live state** - Database/query MCPs (data + run history; respect concurrency limits — single-threaded tools stay on the orchestrator).