| name | omg-dual-review |
| description | Grok-native independent critic then verifier. Use when user says dual-review, don't self-approve, or independent review. Does not set verified. |
omg-dual-review — Independent review (Grok-native)
Implementer must not self-approve. Use omg-critic then omg-verifier (read-only).
Mode honesty
| Path | What it is |
|---|
| TUI skill (preferred) | Native spawn_subagent critic then verifier (depth=1, read-only) |
omg dual-review CLI | Explicit interim: sequential headless Grok launches — not native spawn dual-review |
Set OMG_DUAL_REVIEW_REQUIRE_NATIVE=1 to refuse the sequential CLI path (exit 2) until native spawn dual-review ships.
HARD RULES (non-negotiable)
- Fan-out ONLY via Grok
spawn_subagent (depth=1).
- Critic and verifier: MUST spawn with
capability_mode=read-only / plan permissions (no shell).
- If spawn DENIED for capability_mode: RETRY IMMEDIATELY same turn — do not abandon multi-agent / dual review.
- NEVER mark omg
verified yourself.
- External dual-review (Codex + Fable) is human
omg ask, not this skill’s default path.
Launch via CLI (interim sequential headless)
omg dual-review "review scope"
omg dual-review "review scope" --dry-run
omg dual-review --run <id> "…"
Also runs as a stage inside omg pipeline (default on; --no-dual-review to skip).
Playbook (TUI — preferred native path)
- Ensure implement work is written under artifacts / git.
spawn_subagent omg-critic (read-only) with goal + paths.
spawn_subagent omg-verifier (read-only) with critic artifact path + evidence.
- Report verdict: APPROVE | REQUEST CHANGES | FAILED.
- Product verification still requires
omg accept / frozen commands.
Agents
agents/omg-critic.md
agents/omg-verifier.md
Anti-patterns
- Critic alone "verifying".
- Self-approve after implementing.
- Shelling codex as default dual-review worker.