Back
← All features

Install and operateAvailable in v0.3.0

Commands and next-step messages

Your AI runs oah commands underneath the conversation, guided by the rules oah init installed and each command's next-step message. Every command takes --help, exit codes are fixed (0 done, 1 refused, 2 usage, 69 environment, 75 wait timeout), and every refusal starts with an OAH_* code and says what to run next.

When you need it

When an AI runs tools for you, you need to know what each command did and what should happen next, without memorising a manual.

How it works

  1. Talk in plain words; the installed rules and next-step messages guide your AI through packet, preflight, start, call, gates, review, fix, finish and land.
  2. The command table covers init, doctor, packet, intent, preflight, start/renew/status/finish/release/log, recommend, models, call/wait, gates, check, wiki, impact, eval, graph, review, fix, land, rules, skills, hooks and ci.
  3. oah fix prints one line per next step, starting with the stage it reached; follow them in order.
  4. Ask “Is the review done?” or “Did all the gates pass?” and inspect the evidence before confirming anything.

Example

A result line that names the next command (illustrative review outcome from the worked example).

OAH_REVIEW_CHANGES_REQUESTED: R1 of task hello-name requested changes (P1×1). Correct them (oah fix --task hello-name runs Codex for that), commit, then run oah review --task hello-name again

Requirements

  • oah installed; Claude Code and Codex for the engine stages.

Limits and recovery

  • Exit code 2 also means an ABORT verdict from oah preflight; fix each [abort] line and retry.
  • 75 means oah wait reached its timeout with the run still active; run oah wait <run-id> again.
  • oah eval replays real cases with hidden expected answers and reports each variant, with its limits.

Original documents

Related