review run fails silently as `failed / 0 findings` when codex isn't installed (spawn codex ENOENT)

[vikt0r0]

Summary

ao review run <session> --execute (without an explicit --command) silently fails when the codex CLI is not installed. The review run is recorded as status: failed with 0 open findings and terminationReason: "Codex review failed: spawn codex ENOENT".

This is problematic because:

1. The default reviewer hard-assumes codex is on PATH; there is no preflight check or actionable error at the CLI ("codex not found — install it or pass --command").
2. A failed run with 0 findings is easy to misread as a clean review (no findings) rather than the reviewer never ran. An orchestrator looping over runs can mistakenly treat it as mergeable.
3. There's no config-level way to set a default reviewer command, so every ao review run must pass --command to use a non-codex reviewer.

Environment

• @aoagents/ao 0.9.4
• codex not installed (spawn codex ENOENT)
• Linux

Repro

1. On a host without codex on PATH, open a PR from a worker session.
2. ao review run <session> --execute
3. ao review list <project> → … failed … 0 open findings …
4. Inspect the run JSON → terminationReason: "Codex review failed: spawn codex ENOENT".

Suggestions

• Preflight: detect missing codex before starting and exit with a clear, actionable message (and a non-zero CLI exit).
• Distinguish states: a reviewer that never produced output should not look like failed + 0 findings; consider a distinct errored/reviewer_unavailable status so automation doesn't confuse it with a clean review.
• Configurable default reviewer: support a reviewer.command (or defaults.reviewerCommand) in config.yaml so a custom reviewer can be used without passing --command on every run.
• Optionally document the expected --command stdout contract ({"findings":[{severity,title,body,filePath,startLine,endLine,confidence}]}; {"findings":[]} = clean) in the help text, since it's currently only discoverable from source.