code-combat

To resolve disagreements from an adversarial documentation review, run structured multi-round adversarial negotiation that converges on actionable improvements. The process classifies opponent rebuttals, fact-checks claims, and tracks resolution status until findings converge or require human judgement.

Fact-checking during an AI debate over project documentation works by invoking Codex to verify each opponent's technical claims against official documentation. This prevents incorrect or outdated assertions from carrying through the negotiation rounds, ensuring evidence-based documentation improvements.

A resolution ledger tracks adversarial review findings by maintaining a per-combat record of open, contested, resolved, and conceded items. Stored as ledger.json, it preserves continuity across negotiation rounds so disputed documentation points are systematically tracked until convergence.

Stop an adversarial negotiation over build documentation when convergence rules detect enough P0 and P1 items are resolved or deferred with rationale. Time-boxing rules also halt repeated disputes to require human judgement when progress stalls.

Yes, you can receive adversarial review rebuttal results via Telegram delivery. After classifying opponent responses, fact-checking claims, and updating the resolution ledger, the skill optionally delivers the drafted round response markdown directly through Telegram.

Yes, a prior adversarial-review artifact is required to start an AI documentation debate. The skill applies when a new opponent response references existing findings, loading prior findings and ledger state to fact-check and produce the next round response.