| name | tbm-adversarial-review |
| description | Thin Codex wrapper for TBM adversarial-review prompt construction. Use when LT wants paired attacker/defender prompts for multi-model review loops, attack/defend prompts, arch-rival prompts, or feedback-review loops. Delegates to the canonical repo skill at `C:\Dev\tbm-apps-script\.claude\skills\adversarial-review\SKILL.md`. |
TBM Adversarial Review
This is a Codex adapter, not a second source of truth.
The canonical workflow lives in C:\Dev\tbm-apps-script\.claude\skills\adversarial-review\SKILL.md.
Before acting
- Read
C:\Dev\tbm-apps-script\.claude\skills\adversarial-review\SKILL.md.
- Identify the artifact under review, the defender, the rival, the evidence bar, and the exit condition before generating prompts.
Validation
Before relying on this wrapper, verify this file exists:
C:\Dev\tbm-apps-script\.claude\skills\adversarial-review\SKILL.md
If it is missing, treat the wrapper as stale and say so explicitly instead of guessing.
Adapter rules
- Keep the repo skill canonical for the round protocol and TACT constraints.
- Use this wrapper only to make the workflow discoverable to Codex.
- If the user only wants a single prompt audit rather than a paired attack/defend loop, prefer
tbm-prompt-auditor-pass.
