adversarial-debate-protocol
The adversarial-debate-protocol structures decision-making through three distinct roles: an advocate builds the strongest case for a position, a critic attacks it with rated arguments, and a judge renders verdicts based on explicit evaluation. Use this skill when decisions require rigorous scrutiny to counter confirmation bias, with iteration continuing until the judge accepts the position or rejects it after maximum four rounds of debate.
git clone --depth 1 https://github.com/yogsoth-ai/de-anthropocentric-research-engine /tmp/adversarial-debate-protocol && cp -r /tmp/adversarial-debate-protocol/skills/adversarial-debate-protocol ~/.claude/skills/adversarial-debate-protocolSKILL.md
# Adversarial Debate Protocol A formal three-role debate structure ensuring decisions survive rigorous adversarial challenge. The protocol assigns distinct roles (advocate, critic, judge) to prevent confirmation bias and ensure intellectual honesty. ## Stages 1. **Advocate Construction** — Build the strongest possible case for the position under debate 2. **Critic Attack** — Attack the advocate's case from multiple angles with severity ratings 3. **Judge Verdict** — Impartial assessment of advocate case vs critic attacks, rendering ACCEPT/REJECT/REVISE 4. **Iteration** (if REVISE) — Advocate revises case, critic re-attacks, judge re-evaluates ## Available SOPs | SOP | Role | Purpose | |-----|------|---------| | advocate-construction | Advocate | Build strongest case for position | | critic-attack | Critic | Attack the case with rated arguments | | judge-verdict | Judge | Render impartial verdict | ## Execution Guidance - Minimum 2 rounds before accepting ACCEPT verdict - Critic must produce >= 3 distinct attack arguments per round - Judge must address every critic argument explicitly - If judge verdict is REVISE, advocate must address specific weaknesses identified - Maximum 4 rounds before escalating to strategy level ## Minimum Yield - Advocate case with explicit evidence and reasoning - Critic attacks with severity ratings (HIGH/MEDIUM/LOW) - Judge verdict (ACCEPT/REJECT/REVISE) with point-by-point reasoning - Conditions for acceptance (if ACCEPT) - Required modifications (if REVISE)
Experiment-specific - summarize the DARE executor's research design into a clean research_result report, forced to write back into the spec file produced by formated-specs.
Experiment-specific - replaces writing-specs, emits DARE's 4-layer call plan as a clean research_graph schema. Last step forces load formated-result.
loss-1 judge - read a sample's full dialogue and decide whether the user simulator semantically enacted its Policy Card. check-blind.
loss-2 judge - pairwise quality comparison across the n rungs within one topic; decide monotonicity and endpoint separation. check-blind, D1-D5 only.
Strategy: 面对异常的最佳解释推理
Remove components one by one, observe system changes to reveal hidden dependencies and generate ideas from structural gaps.
Map system architecture to ablatable units for ablation studies
Design ablation studies to isolate component contributions in ML systems