Claude Grep — Development skill for Claude Code

claude-grep

[](https://github.com/evoleinik/claude-grep/actions/workflows/ci.yml) [](https://app.codecov.io/gh/evoleinik/claude-grep) [](https://goreportcard.com/report/github.com/evoleinik/claude-grep) [](https://pkg.go.dev/github.com/evoleinik/claude-grep)

grep for your Claude Code conversations.

Every Claude Code session is stored as JSONL under `~/.claude/projects/`. That's a searchable record of every problem you've solved, every gotcha you've hit, and every decision you've made — but it's invisible. `claude-grep` makes it greppable: regex or semantic (vector) search over your entire history, with output tuned for both humans at a terminal and agents with a finite context window.

It's a single Go binary with no runtime dependencies (ollama is optional, only for semantic search).

Why

Problem	claude-grep
Sessions are buried in JSONL files	searches them like grep
"Where's that conversation about X?"	-s finds it by meaning, not exact words
Python tooling adds ~200ms startup	Go binary starts in <10ms
Raw matches flood the context window	BM25 compression returns only query-relevant text
No structured output for piping	--json on every command
Empty results give an agent zero signal	prints scope + next-step suggestions on a miss

Install

go install github.com/evoleinik/claude-grep@latest

Or build from source:

git clone https://github.com/evoleinik/claude-grep.git
cd claude-grep
go build -o claude-grep .

**Semantic search** is optional and needs [ollama](https://ollama.com) with the `nomic-embed-text` model:

curl -fsSL https