GoKawiilGrepathy
Make agent-written code reviewable.
When a coding agent builds a feature, it makes a bunch of small decisions along the way that you never really approved. Later someone reviews the PR and asks "why was this done this way?" and you don't know. The answer is buried in a chat transcript on your laptop, and Claude Code deletes those after 30 days by default.
Grepathy reads your session transcripts locally, pulls out the decisions, and writes them to a markdown file that gets committed with your code. Now the repo can answer "why" on its own, for reviewers, teammates, and future agents.
npx grepathy init
What it looks like
Real example: on a contract project, my agent decided on its own to pre-create guest users in Clerk. It wasn't in any plan. The CTO saw it in the PR and asked me why. I had no idea, because I hadn't made that decision. With Grepathy, it would have been in the repo:
### Guest identities are pre-created in Clerk Status: agent-initiated — not requested in plan or prompts Touches: ` lib/clerk/* ` , ` db/schema/guests.ts ` The agent inferred this approach to simplify downstream auth checks. No explicit rationale was discussed. Risk: guest users diverge from the normal signup path. Reviewer attention: confirm whether guests should be modeled as normal users.
Run grep -rn "agent-initiated" .ai/why/ and you get a list of every decision the agent made without asking anyone.
How it works
... continue reading