Skip to content
Code Archaeology Read the codebase like a book. Argue with every chapter.

Other agents

Claude Code, OpenCode, and Cursor get plug-and-play install. Other agents need a one-time copy-paste because they don’t have a skills directory.

The pattern is the same everywhere:

  1. Copy the contents of dig/SKILL.md into your agent’s long-lived instruction file (AGENTS.md, ~/.aider.conf.yml, system prompt, etc.).
  2. Tell the agent “follow the dig skill” or “do an excavation pass” — the skill instructions do the rest.

Codex CLI

Section titled “Codex CLI”

Codex reads AGENTS.md from the repo root and from ~/.codex/AGENTS.md.

Terminal window
mkdir -p ~/.codex
cat ~/.local/share/indiana-codes/dig/SKILL.md >> ~/.codex/AGENTS.md

Or, if you’d rather not pollute every conversation, paste it inline when you want to dig:

Follow these instructions to excavate this codebase:


<paste contents of dig/SKILL.md>

Aider

Section titled “Aider”

Aider supports a config-level read: list for files always loaded into context.

~/.aider.conf.yml
read:
  - ~/.local/share/indiana-codes/dig/SKILL.md

Start aider and say “follow the dig skill”.

Continue.dev

Section titled “Continue.dev”

Add a custom slash command in ~/.continue/config.yaml:

customCommands:
  - name: dig
    description: Code archaeology — chapter-based history walkthrough
    prompt: |
      <paste contents of dig/SKILL.md>

Zed AI

Section titled “Zed AI”

Drop the contents into a .rules file at the repo root, or paste into the assistant panel as a system message. Zed honours .rules automatically when its assistant runs.

Generic / any chat-style assistant

Section titled “Generic / any chat-style assistant”

The skill is self-contained. Paste dig/SKILL.md into the system prompt of whatever assistant you’re using and say “begin Phase 1.” Works in raw Claude / GPT / Gemini chat windows too — you just lose the slash-command ergonomics and have to run git commands yourself when the assistant asks for output.

A note on tool access

Section titled “A note on tool access”

Code Archaeology gets much better when the agent can call git, gh, and any issue-tracker MCPs you have configured. Agents that can’t shell out (most chat sandboxes) will stall at Phase 1 — you’ll have to be the hands-and-feet, pasting command output back to the model. Workable, but slow. Prefer an agent with shell access if you have the choice.