llvm-harness

Look up how to use LLVM command-line tools and utilities from the build directory, as well as where to find useful LLVM materials. Use this when you need to know which binary to run, what flags to pass, or how to interpret output from opt, llc, lli, llvm-lit, alive-tv, and other LLVM tools — even if the user just says "run the tests" or "check if this is correct" without naming a specific tool. Use this when you need to find more details about a specific LLVM entity (e.g. a pass name, a debug flag, a FileCheck directive) or when you need to find documentation on a specific topic (e.g. IR semantics, pass design, loop terminology).

Curate raw insights recorded during an agent run. Reviews insights in local/ scopes, evaluates each for generalizability and novelty, then promotes worthy ones to shared/ scopes. Call this after a successful fix to capture lasting knowledge.

Search and synthesize LLVM insights using natural language. Use this when keyword_search is too narrow or when asking a conceptual question like "What are common pitfalls when folding binary operators in InstCombine?"

Review a patch for fixing an LLVM bug and produce a structured report that identifies, categorizes, and prioritizes potential issues in the patch. Use this skill when you have a patch that has passed LLVM testing and is ready for review. Pass the previously reviewed patch and the review report as `argument`, if present, so the reviewer can check for regressions from earlier attempts.