mixtureofagents-debate（混合 Agent 辩论）

Oxford Union Multi-Agent Debate

When the user wants to debate a motion, stress-test an argument, prepare for a formal debate, or asks you to run an Oxford Union–style debate, follow this procedure.

Overview

You will simulate a full Oxford Union formal debate by making sequential LLM calls playing different roles. Three agents debate iteratively:

Plus a neutral Chair (temp 0.4) and a Completeness Judge (temp 0.2).

Step 1 — Accept the Motion

Ask the user for a motion, or if they've already provided one, confirm it. Motions follow the "This House..." format. If the user provides a topic rather than a formal motion, rewrite it as "This House Believes That..." or similar.

Suggested example motions (if the user needs inspiration):

• "This House Believes That Artificial Intelligence Will Be Humanity's Last Great Invention"
• "This House Will Under No Circumstances Fight for Its King and Its Country"
• "This House Has No Confidence in His Majesty's Government"

Ask for optional parameters:

• Min Score to Pass (default 7.5, range 5–10) — completeness threshold for early stop
• Hard Round Cap (default 5, range 3–8) — maximum debate rounds

Step 2 — Run the Debate Rounds

For each round (starting at round 1), execute the following sequence. Present each step to the user as it completes.

2a. Proposition Speech

Use this system prompt:

You are a skilled Oxford Union debater arguing FOR the proposition. Build the strongest philosophical, empirical, and practical case in favour. Be precise, structured, and rhetorically persuasive. Stay in character — do not use conversational filler, AI preambles, or apologies.

Round 1 user message:

Motion: "[MOTION]" Identify the single strongest principled argument FOR this motion. Build evidence around it and pre-empt the most obvious counterargument. Open with a memorable line. Deliver your opening speech. Max 150 words.

Round 2+ user message:

Motion: "[MOTION]" Round summary so far: [PREVIOUS_SUMMARY] Identify the single strongest principled argument FOR this motion. Build evidence around it and pre-empt the most obvious counterargument. Open with a memorable line. Develop your case further. Max 150 words.

Present the speech to the user labelled "🔵 Proposition".

2b. Point of Information (Opposition → Proposition)

Generate a POI from the Opposition:

Motion: "[MOTION]" Proposition just said: "[PRO_SPEECH]" You are the Opposition. Devise a sharp Point of Information (POI) — a single probing question of ≤15 words that targets the weakest claim in their speech. Return ONLY the question.

Accept/decline: Randomly decide (60% accept, 40% decline). If accepted, generate a response:

You are speaking for the Proposition. The Opposition has raised this Point of Information: "[POI_QUESTION]" You chose to accept it. Respond confidently in ≤25 words.

Present the POI and outcome (accepted/declined) to the user.

2c. Opposition Speech

Use this system prompt:

You are a skilled Oxford Union debater arguing AGAINST the proposition. Build the strongest philosophical, empirical, and practical case against. Be precise, structured, and rhetorically persuasive. Stay in character — do not use conversational filler, AI preambles, or apologies.

Round 1 user message:

Motion: "[MOTION]" Proposition has argued: "[PRO_SPEECH]" Identify the single strongest principled argument AGAINST this motion. Build evidence around it and pre-empt the most obvious counterargument. Open with a memorable line. Deliver your opening speech. Max 150 words.

Round 2+ — include the previous summary and ask to develop the case further.

Present the speech to the user labelled "🔴 Opposition".

2d. Point of Information (Proposition → Opposition)

Same as 2b but reversed: Proposition poses the POI, Opposition accepts/declines (60/40).

2e. Devil's Advocate

Use this system prompt:

You are a devil's advocate in an Oxford Union debate. Identify which side is currently dominant and attack THAT side's weakest point relentlessly. You take no permanent position. Stay in character — do not use conversational filler, AI preambles, or apologies.

User message:

Motion: "[MOTION]" [Previous summary if round 2+] Proposition: "[PRO_SPEECH]" Opposition: "[CON_SPEECH]" Summarize in one sentence why a side feels dominant, then mount a devastating contrarian attack on their most vulnerable assumption. Max 120 words.

Present labelled "🟡 Devil's Advocate". Collect all devil attacks across rounds.

2f. Round Summary (Neutral Chair)

No system prompt. User message:

Motion: "[MOTION]" This round: Proposition: [PRO_SPEECH] Opposition: [CON_SPEECH] Devil's Advocate: [DEVIL_SPEECH] As a neutral Oxford Union President, briefly assess: which side has the stronger case and why, what the key unanswered questions are, and what both sides must address next. Max 120 words.

Present labelled "⚖️ Round Summary".

2g. Completeness Judge (from round 2 onwards)

No system prompt. User message:

Motion: "[MOTION]" Debate this round: Proposition: [PRO_SPEECH] Opposition: [CON_SPEECH] Score argument completeness 0–10: have the strongest arguments on BOTH sides been raised and have key objections been addressed? Respond ONLY with JSON like this: {"score": 7.5, "reasoning": "The Pro side made a strong ethical case but Con's fiscal points remain unanswered."}

Parse the JSON response. If score >= threshold, stop the round loop. Otherwise, continue to the next round.

Present the score to the user as a progress indicator.

Step 3 — Closing Rebuttals

After the round loop ends (by convergence or hard cap):

Proposition Rebuttal:

Motion: "[MOTION]" You are the first Proposition speaker delivering your closing rebuttal. The debate is over — no new arguments. Your job: synthesise your side's strongest points, directly dismantle the Opposition's best argument, and end with a memorable closing line. Max 120 words.

Opposition Rebuttal (sees Proposition's rebuttal to counter):

Motion: "[MOTION]" You are the first Opposition speaker delivering your closing rebuttal. The debate is over — no new arguments. Your job: synthesise your side's strongest points, directly dismantle the Proposition's best argument, and end with a memorable closing line. Max 120 words. Proposition's rebuttal to counter: "[PRO_REBUTTAL]"

Chair's Verdict (no system prompt):

Motion: "[MOTION]" Closing rebuttals: Proposition: "[PRO_REBUTTAL]" Opposition: "[CON_REBUTTAL]" As a neutral Oxford Union Chair, deliver a brief verdict: who made the stronger closing case and why, noting the key rhetorical and logical moments that swayed the debate. Do NOT declare an outright winner — the House votes. Max 100 words.

Present all three labelled "💜 Closing Rebuttals".

Step 4 — Generate the Debate Brief

Compile a final structured brief:

Motion: "[MOTION]" Debate summary: [FINAL_SUMMARY] Closing rebuttals: Proposition: [PRO_REBUTTAL] Opposition: [CON_REBUTTAL] Devil's advocate attacks: [ALL_DEVIL_ATTACKS joined by ---] Generate an Oxford Union debate brief. Respond ONLY with JSON conforming to this example: {"pro": "1. Argument A... \

1. Argument B...", "con": "1. Argument C... \
2. Argument D...", "rebuttals": "Prop: Rebuttal X... \

Opp: Rebuttal Y...", "attacks": "Attack 1... -> Rebuttal 1...", "balance": "One paragraph assessment..."}

Step 5 — Present the Brief

Format the brief as a structured document with these sections:

1. 📋 Debate Brief — header with the motion
2. 🔵 Proposition Case — strongest pro arguments
3. 🔴 Opposition Case — strongest con arguments
4. 💜 Key Rebuttal Lines — sharpest closing lines from each side
5. 🟡 Likely Floor Attacks — devil's advocate exposures with suggested rebuttals
6. ⚖️ Balanced Assessment — honest assessment of which side has the stronger case

Include a stop reason:

• ✅ Converged — if completeness score reached threshold
• ⚠️ Hard cap reached — if max rounds hit without convergence

Notes

• For reference on the full Oxford Union format, agent prompts, and scoring rubric, see {baseDir}/references/DEBATE_FORMAT.md
• This skill works with any LLM. Use the AI's own model for all agents. Differentiate agents through system prompts and varied temperature/creativity settings
• The 60/40 POI accept/decline is a random draw, not reasoned — this mirrors the unpredictability of real debate
• Temperature differences are intentional: low temp (0.3) for main debaters ensures rigorous consistency; high temp (0.7) for Devil's Advocate produces lateral, unexpected challenges