Research
Answer a hard, open-ended question about how the project works or is built by synthesizing every source — code, structure, git history, prior decisions, discussion, caveats — and verifying each claim against the source. Use for deep cross-cutting questions that span more than one area. Narrower siblings — a single area's structure is explore, the rationale behind one past decision is recall, a failure's root cause is debug.From its SKILL.md
npx -y skills add TheWinci/mimirs --skill researchAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
2.1 KB, as published. Nobody here has run it
Research
Goal: a thorough, cited answer to a specific question — pulled from across the whole project memory, not one file or one area. Where explore maps an area's structure, research answers a question by combining code, history, decisions, and discussion.
- Frame the question — break it into the sub-questions you must answer. State them, so the synthesis can be checked against them at the end.
- The code —
search "<sub-question>"for the relevant files, thenread_relevant "<behavior>"for the actual functions/sections with line ranges. Use theextensions/dirs/excludeDirsfilters to scope.search_symbols <name>for named things you already know. - How it connects — when the question is about relationships:
project_map(focus: <file>),depends_on/dependents <file>,impact <symbol>(transitive callers),trace from=<a> to=<b>(how one reaches another). - History & why —
search_commits "<topic>"for why and when it changed;file_history <file>for how a file evolved. - Decisions & discussion —
search_checkpoints "<topic>"for recorded decisions and direction changes;search_conversation "<topic>"for the back-and-forth and alternatives considered. - Caveats —
get_annotations(search across notes) for known bugs, constraints, and fragile spots that qualify the answer. - Verify — before asserting any claim, reopen the cited source and confirm it. Don't synthesize from snippets alone.
Finish: answer each sub-question in plain language with a file:line, commit, or checkpoint citation; separate what the source proves from what is inferred; list what stayed unresolved instead of guessing.