Optimize by AI Agents

Run an end-to-end optimization session driven by an AI agent: install the Claude Code  slash commands into your project, scope a measurable target with /codespine-interview, and apply a verified-safe edit with /codespine-optimize. The agent runtime is Claude Code itself, so there is no API key or LLM provider to configure.

Setup

1. Open the project you want to optimize. It must be a TypeScript project with a tsconfig.json — see the prerequisites.

2. Install the codespine assets into the project’s .claude/ directory. Run this from the root of the repository; it creates .claude/ for you (see install for options):

   npx codespine install

3. Start Claude Code from the root of the repository:

   claude

4. Scope a target. Launch the interview — it is read-only and asks you questions to pin down a measurable optimization target, grounding each candidate in the graph:

   /codespine-interview

5. Apply an optimization. The interview often proposes this command for you, but you can also run it on its own with an explicit target:

   /codespine-optimize
   /codespine-optimize the database FooBar needs an index

See Agent slash commands for what each command does.

Keep a record of the session

A session leaves a git diff behind, but it helps to capture the reasoning too:

• Open a GitHub issue about the session. Ask Claude directly — for example, “create a GitHub issue about this optimization session” — so the what and why are tracked alongside the diff.

• Export the transcript to markdown with Claude Code’s /export command:

  /export /path/to/session.md

See also

• Agent slash commands — what each /codespine-* command does.
• The pipeline — extract → load → query → run the agent on its first verified edit.
• install — the command that drops the slash commands into .claude/.