Skip to content

The skill pack

Every workspace comes with written instructions for Claude Code and Codex, so they use your case the way Area O1 expects.

What's in your workspace

File For What it is
.claude/skills/areao1/SKILL.md Claude Code A skill named areao1. Claude Code loads it when you ask about your case, criteria, gaps, evidence, deadlines or letters, or share something that matters for the case.
AGENTS.md Codex and other agents that read it The workspace guide: rules, the folder layout, useful commands, and a section called "Driving Area O1 from Claude Code or Codex".

Both are plain Markdown. You can read them in any editor. The originals ship with Area O1 in areao1/templates/workspace/.

What they tell agents

Both files give the same rules. In short:

  • Connect through the MCP server. The setup line (claude mcp add ... or codex mcp add ...) and the six tools, with what each one does.
  • Start each session with what_changed(since=<last session>) and get_scoreboard().
  • Check before relying on a fact. Use get_provenance. Draft only from claims whose status is approved and that are current. Where proof is missing, say what's unknown.
  • Offer to save what matters. When you share a deadline, an invitation, a result or a letter writer, the agent offers to send it with propose_context, in your words, with its own name as client.
  • An invitation is not a completion, and a preprint is not a publication. Respect each claim's stage.
  • No verdicts. Never say you qualify, will be approved or meet a criterion. Any judgment is an opinion and must be labeled as one.
  • Don't touch evidence or memory files. Never edit data/exhibits.json, evidence/ or memory/*.jsonl directly, and never delete anything. Evidence enters only through the Inbox, accepted by you.
  • No secrets in files. Keys and tokens live in the OS keychain.
  • Never send email for you. Area O1 only sends an email you approved on the Contacts page.

AGENTS.md adds the folder layout (what lives in data/, evidence/, memory/, drafts/letters/), the file naming rule for exhibits, and a few commands (areao1 up, areao1 run sync, areao1 validate and others). It also tells agents to run areao1 validate after editing a data/*.json file and not to edit generated files such as DASHBOARD.md and data/criteria.json.

The MCP server sends the agent a shorter version of the same rules when it connects, so clients that never read these files still get them.

When they're installed

  • areao1 init (and the first-run setup) writes both files into a new workspace.
  • Each time the app starts (areao1 up), Area O1 checks the workspace. If SKILL.md is missing, it adds it. If AGENTS.md is missing, it writes it. If AGENTS.md exists but has no "Driving Area O1 from Claude Code or Codex" section, it appends that section.

It only adds what's missing. It never rewrites a file you already have, so your own edits to AGENTS.md stay.

Refresh them after an update

Because existing files are never overwritten, a newer version of Area O1 doesn't replace your copies. To get the current ones:

  1. Stop the app.
  2. To refresh the skill, delete .claude/skills/areao1/SKILL.md.
  3. To refresh the Claude Code and Codex section of AGENTS.md, delete that section (from its heading to the next ## heading). To refresh the whole guide, delete AGENTS.md. Copy out any notes of your own first.
  4. Start the app again with areao1 up. The missing pieces are written from the installed version.

Add your own notes

You can add your own sections to AGENTS.md, for example how you like letters drafted. Keep the rules section as it is: it's what keeps agents from editing evidence directly.