agent-debate

Name: agent-debate
Author: WenyuChiou

16stars4forks

How to install agent-debate?

wenyuchiou/agent-collab-skills/agent-debate

$npx -y skills add WenyuChiou/agent-collab-skills --skill agent-debate

Installs into the current project.

Summary

Use when a consequential decision needs adversarial review, opposing agent arguments, a second opinion via debate, or explicit trade-off analysis before implementation.

Files · 1

View on GitHub

SKILL.md

1	# agent-debate
2
3	Anti-consensus tool. Most LLM output collapses to a single
4	"reasonable-sounding" answer that hides real trade-offs. For
5	consequential decisions (you'll regret picking the wrong
6	side), this skill stages an explicit pro/con debate between two
7	agents, then makes the disagreement legible.
8
9	This is not for routine choices — variable naming, file layout,
10	which library to use for a one-off script. It's for decisions where
11	there's a real trade-off and the "obvious" answer might be wrong.
12
13	## When to use
14
15	Trigger phrases:
16
17	- "Have Claude and Codex debate this design choice."
18	- "Adversarial review on `<decision>`."
19	- "Get a second opinion via debate."
20	- "I'm not sure about this — make the agents argue it out."
21	- "Steel-man both sides on `<topic>`."
22
23	Not for:
24
25	- Routine choices (pick whatever; doesn't matter long-term).
26	- Decisions where you already know the right answer and just want
27	validation (that's confirmation bias; the debate will waste
28	tokens).
29	- Open-ended brainstorming (use a single agent + lots of follow-ups).
30
31	## Inputs
32
33	User must provide:
34
35	1. The decision — phrased as a yes/no or A/B question.
36	- Good: "Should we keep legacy.py as a deprecation shim or
37	remove it?"
38	- Bad: "What should we do about legacy.py?" (too open)
39	2. Optional: which agent argues which side. If unspecified, the
40	skill picks based on agent strengths (see below).
41	3. Optional: number of rounds. Default 2 (each agent gets 2
42	turns). More than 3 rarely adds signal.
43	4. Optional: context policy. Default debate turns are 200-400
44	words, final synthesis is <= 250 words, and only the accepted
45	decision is eligible for shared memory.
46
47	## Workflow
48
49	### 1. Frame the decision
50
51	Restate the decision as a clear A vs B (or yes vs no). If the user
52	gave an open-ended question, propose the framing back to them and
53	get confirmation before spending tokens on a debate.
54
55	Example:
56	- User: "Should we cache LLM responses?"
57	- Skill: "Framing: **A) Yes, add an LRU cache keyed on prompt
58	hash vs B) No, every call is unique enough that caching adds
59	bug surface without saving money.** Confirm before I proceed."
60
61	### 2. Assign sides
62
63	Default heuristic:
64
65	\| Position character \| Best argued by \|
66	\|---\|---\|
67	\| "We should ship the simpler / less-surface option" \| Claude (judgment-heavy, conservative bias) \|
68	\| "We should optimize for X performance / scale dimension" \| Codex (concrete with numbers) \|
69	\| "We should consider edge cases X / Y / Z that the obvious answer misses" \| Gemini (long-context, edge-case catching) \|
70	\| "We should follow the convention from <reference codebase or paper>" \| Gemini (synthesizing from external sources) \|
71
72	If the user explicitly assigns sides, use those. Otherwise propose
73	the assignment and confirm.
74
75	### 3. Initialize `.coord/debate_<topic>.md`
76
77	```markdown
78	# Debate — <topic-slug>
79
80	Decision: <one-line A vs B framing>
81	Rounds: 2
82	Side A (argued by Codex): <position A>
83	Side B (argued by Gemini): <position B>
84	Started: 2026-04-28T...
85
86	---
87
88	## Round 1
89
90	### Side A — Codex
91	<task: read .coord/debate_<topic>.md, argue Side A from scratch>
92	<argument here>
93
94	### Side B — Gemini
95	<task: read .coord/debate_<topic>.md (which now contains Codex's
96	arg), argue Side B as a rebuttal>
97	<argument here>
98
99	---
100
101	## Round 2
102
103	### Side A — Codex
104	<task: read updated debate file, rebut Side B>
105	...
106	```
107
108	### 4. Generate task files for the debate rounds
109
110	For each agent's turn, write a task file in their delegate-skill
111	format:
112
113	`.ai/codex_task_debate_<topic>_round1.md`:
114
115	```markdown
116	# Task: Argue Side A in debate on <topic>
117
118	## Context
119	- Repo: <path>
120	- Read: .coord/debate_<topic>.md (the debate transcript so far)
121	- Side A: <position A>
122	- Your role: argue Side A. Be specific. Reference real code / files
123	/ numbers. Don't hedge ("on the other hand" is forbidden).
124	- Do NOT touch any source files. This is argument-only.
125
126	## Goal
127	Write a 200-400 word argument for Side A, focused on:
128	- The strongest specific reason Side A is right.
129	- The most concrete failure mode of Side B (with a worked example
130	if possible).
131	- One thing Side B might say that you concede is a real cost of
132	Side A.
133
134	## Acceptance
135	- Append your argument under "## Round 1 / ### Side A — Codex" in
136	.coord/debate_<topic>.md.
137	- Keep argument 200-400 words. Longer = penalty.
138	- No "on the other hand" hedging.
139	```
140
141	Subsequent rounds: same template, but task file says "rebut Side B's
142	Round N argument; concede any genuinely strong point but explain
143	why Side A still wins net." Each round's task file references the
144	previous rounds in the debate file as input.
145
146	### 5. Run the rounds
147
148	Hand off to `codex-delegate` / `gemini-delegate` to execute each
149	agent's turn. They write into the debate file directly (the task
150	file says where to append).

People who liked this also viewed

6 picks

Type

lark-openapi-explorer飞书/Lark 原生 OpenAPI 探索：从官方文档库中挖掘未经 CLI 封装的原生 OpenAPI 接口。当用户的需求无法被现有 lark-* skill 或 lark-cli 已注册命令满足，需要查找并调用原生飞书 OpenAPI 时使用。SkillsJul 2026378k16k
lark-skill-maker创建 lark-cli 的自定义 Skill。当用户需要把飞书 API 操作封装成可复用的 Skill（包装原子 API 或编排多步流程）时使用。SkillsJul 2026377k16k
azure-cost-optimizationOfficial agent plugin providing skills and MCP server configurations for Azure scenarios.SkillsJul 2026209k1.3k
vercel-react-native-skillsVercel's official collection of agent skillsSkillsJul 2026172k29k
mcp-builderGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate externaSkillsJul 202693k163k
gws-sharedgws CLI: Shared patterns for authentication, global flags, and output formatting.SkillsJul 202653k30k

agent-debate

16stars4forks

View on GitHub

How to install agent-debate?

wenyuchiou/agent-collab-skills/agent-debate

$npx -y skills add WenyuChiou/agent-collab-skills --skill agent-debate

Installs into the current project.

Summary

Use when a consequential decision needs adversarial review, opposing agent arguments, a second opinion via debate, or explicit trade-off analysis before implementation.

Files · 1

View on GitHub

SKILL.md

1	# agent-debate
2
3	Anti-consensus tool. Most LLM output collapses to a single
4	"reasonable-sounding" answer that hides real trade-offs. For
5	consequential decisions (you'll regret picking the wrong
6	side), this skill stages an explicit pro/con debate between two
7	agents, then makes the disagreement legible.
8
9	This is not for routine choices — variable naming, file layout,
10	which library to use for a one-off script. It's for decisions where
11	there's a real trade-off and the "obvious" answer might be wrong.
12
13	## When to use
14
15	Trigger phrases:
16
17	- "Have Claude and Codex debate this design choice."
18	- "Adversarial review on `<decision>`."
19	- "Get a second opinion via debate."
20	- "I'm not sure about this — make the agents argue it out."
21	- "Steel-man both sides on `<topic>`."
22
23	Not for:
24
25	- Routine choices (pick whatever; doesn't matter long-term).
26	- Decisions where you already know the right answer and just want
27	validation (that's confirmation bias; the debate will waste
28	tokens).
29	- Open-ended brainstorming (use a single agent + lots of follow-ups).
30
31	## Inputs
32
33	User must provide:
34
35	1. The decision — phrased as a yes/no or A/B question.
36	- Good: "Should we keep legacy.py as a deprecation shim or
37	remove it?"
38	- Bad: "What should we do about legacy.py?" (too open)
39	2. Optional: which agent argues which side. If unspecified, the
40	skill picks based on agent strengths (see below).
41	3. Optional: number of rounds. Default 2 (each agent gets 2
42	turns). More than 3 rarely adds signal.
43	4. Optional: context policy. Default debate turns are 200-400
44	words, final synthesis is <= 250 words, and only the accepted
45	decision is eligible for shared memory.
46
47	## Workflow
48
49	### 1. Frame the decision
50
51	Restate the decision as a clear A vs B (or yes vs no). If the user
52	gave an open-ended question, propose the framing back to them and
53	get confirmation before spending tokens on a debate.
54
55	Example:
56	- User: "Should we cache LLM responses?"
57	- Skill: "Framing: **A) Yes, add an LRU cache keyed on prompt
58	hash vs B) No, every call is unique enough that caching adds
59	bug surface without saving money.** Confirm before I proceed."
60
61	### 2. Assign sides
62
63	Default heuristic:
64
65	\| Position character \| Best argued by \|
66	\|---\|---\|
67	\| "We should ship the simpler / less-surface option" \| Claude (judgment-heavy, conservative bias) \|
68	\| "We should optimize for X performance / scale dimension" \| Codex (concrete with numbers) \|
69	\| "We should consider edge cases X / Y / Z that the obvious answer misses" \| Gemini (long-context, edge-case catching) \|
70	\| "We should follow the convention from <reference codebase or paper>" \| Gemini (synthesizing from external sources) \|
71
72	If the user explicitly assigns sides, use those. Otherwise propose
73	the assignment and confirm.
74
75	### 3. Initialize `.coord/debate_<topic>.md`
76
77	```markdown
78	# Debate — <topic-slug>
79
80	Decision: <one-line A vs B framing>
81	Rounds: 2
82	Side A (argued by Codex): <position A>
83	Side B (argued by Gemini): <position B>
84	Started: 2026-04-28T...
85
86	---
87
88	## Round 1
89
90	### Side A — Codex
91	<task: read .coord/debate_<topic>.md, argue Side A from scratch>
92	<argument here>
93
94	### Side B — Gemini
95	<task: read .coord/debate_<topic>.md (which now contains Codex's
96	arg), argue Side B as a rebuttal>
97	<argument here>
98
99	---
100
101	## Round 2
102
103	### Side A — Codex
104	<task: read updated debate file, rebut Side B>
105	...
106	```
107
108	### 4. Generate task files for the debate rounds
109
110	For each agent's turn, write a task file in their delegate-skill
111	format:
112
113	`.ai/codex_task_debate_<topic>_round1.md`:
114
115	```markdown
116	# Task: Argue Side A in debate on <topic>
117
118	## Context
119	- Repo: <path>
120	- Read: .coord/debate_<topic>.md (the debate transcript so far)
121	- Side A: <position A>
122	- Your role: argue Side A. Be specific. Reference real code / files
123	/ numbers. Don't hedge ("on the other hand" is forbidden).
124	- Do NOT touch any source files. This is argument-only.
125
126	## Goal
127	Write a 200-400 word argument for Side A, focused on:
128	- The strongest specific reason Side A is right.
129	- The most concrete failure mode of Side B (with a worked example
130	if possible).
131	- One thing Side B might say that you concede is a real cost of
132	Side A.
133
134	## Acceptance
135	- Append your argument under "## Round 1 / ### Side A — Codex" in
136	.coord/debate_<topic>.md.
137	- Keep argument 200-400 words. Longer = penalty.
138	- No "on the other hand" hedging.
139	```
140
141	Subsequent rounds: same template, but task file says "rebut Side B's
142	Round N argument; concede any genuinely strong point but explain
143	why Side A still wins net." Each round's task file references the
144	previous rounds in the debate file as input.
145
146	### 5. Run the rounds
147
148	Hand off to `codex-delegate` / `gemini-delegate` to execute each
149	agent's turn. They write into the debate file directly (the task
150	file says where to append).

People who liked this also viewed

6 picks

Type