Agentgit — Git skill for Claude Code

agentgit

See the context behind every line of code written by coding agents.

After a session with Claude Code or Codex, you can diff to see *what* changed—but not *why*. Which prompt triggered that refactor? What was the agent thinking when it modified that line?

agentgit transforms raw agent transcripts into a **structured transcript**—a separate git history that preserves the full story without touching your codebase.

**Each coding session becomes its own branch.** Each user prompt becomes a merge commit, with the agent's individual turns (thinking + changes) as child commits:

Your agentgit repo:

  main
   │
   └─── session/claude-code/add-user-auth
   │     │
   │     ├─ [MERGE] Prompt: "Add user authentication"
   │     │   ├─ Create auth.py with JWT utilities
   │     │   │  Context: I'll implement JWT token generation...
   │     │   ├─ Add authentication middleware
   │     │   │  Context: Creating middleware to verify tokens...
   │     │   └─ Update login component with auth flow
   │     │      Context: Integrating the auth into the UI...
   │     │
   │     └─ [MERGE] Prompt: "Add password reset"
   │         ├─ Create reset token generator
   │         └─ Add reset email template
   │
   └─── session/claude-code/fix-database-bugs
   │     │
   │     └─ [MERGE] Prompt: "Fix connection timeout"
   │         ├─ Debug connection pool settings
   │         │  Context: Found the pool size was too small...
   │         └─ Add retry logic with exponential backoff
   │            Context: Implementing retries to handle transient failures...
   │
   └─── session/codex/refactor-api
         │
         └─ [MERGE] Prompt: "Refactor payment endpoints"
             ├─ Extract payment handlers to separate module
             ├─ Consolidate error handling
             └─ Add request validation middleware

Each commit preserves the agent's reasoning and the full context:

$ agentgit log session/claude-code/add-user-auth --oneline
a1b2c3d Update login c