Repo graph debug
Start from a failure, not a filename — turn a stacktrace, failing test, or diff into the exact code that matters, then read it. Use when you have an error and need to know where to look.From its SKILL.md
npx -y skills add James-Chahwan/repo-graph --skill repo-graph-debugAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
SKILL.md
1.4 KB, as published. Nobody here has run it
repo-graph debug
Paste the error, get the code. The on-ramp when something's broken — instead of grepping file names out of a traceback and opening each one.
When to use
- You have a stacktrace, a failing-test id (
path::test_name), or a diff / list of changed files. - "Where do I even start with this error?"
Steps
- Call
findwithquery=<the raw stacktrace / test id / unified diff>. It sniffs the shape (or passkind=stacktrace|test|diff), resolves the frames/symbols/paths to nodes, then ranks the surrounding subgraph by relevance (Personalized PageRank). Every row carriespath:line. - Call
readwithnode=<top result>to pull that node's exact source, sliced from its file by the graph's line span — no grep, no scrolling. Addcontext_lines=Nfor padding around it. - Widen if needed:
impacton the node for blast radius (it takes several comma-separated nodes — feed it the whole changed set) —direction=bothatdepth=1for direct callers/callees.
find only resolves frames that actually exist in the graph — if it returns nothing, the signal's paths/symbols don't match this repo; fall back to find with a plain keyword.