evidence-redteam

Validates claims by gathering confirming and disconfirming evidence, then red-teams the reasoning.

Updated Jun 28, 2026
One-click install
npx skills add https://github.com/Cohey0727/CodingAgentTools --skill evidence-redteam-cohey0727
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: evidence-redteam
Source: https://github.com/Cohey0727/CodingAgentTools/tree/main/skills/evidence-redteam
Command: npx skills add https://github.com/Cohey0727/CodingAgentTools --skill evidence-redteam-cohey0727

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When you need to defend a claim, plan, or decision, it is easy to collect only supporting evidence and miss the weaknesses that critics will attack. This Skill investigates both confirmation and disconfirmation in parallel, stress-tests the logic itself, and prepares rebuttal material before anyone challenges you. ## Core Features & Use Cases - Two-sided evidence gathering: Runs parallel research agents (confirmation team, disconfirmation team, base-rate team, expert-consensus team) using WebSearch, WebFetch, and optionally Parallel Search MCP, with mandatory source URLs and cross-verification. - Red-team critique with independent second opinion: Attacks hidden assumptions, logical gaps, alternative hypotheses, and failure modes, then sends the case to a separate LLM via codex exec for an adversarial review whose findings are re-verified before adoption. - Pre-mortem mitigation report: Ranks anticipated criticisms by severity and likelihood, pairing each with rebuttal evidence, mitigation strategies, or honestly acknowledged limitations. - Use Case: Before proposing "microservices will speed up development" to leadership, run this Skill to get a verdict on validity, a sourced evidence table, the strongest counterarguments, and prepared responses for the meeting. ## Quick Start Ask the agent to run evidence-redteam on your claim, for example: validate the claim that our new product will grow 20% annually in the domestic market.

Frequently Asked Questions about evidence-redteam

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I critically evaluate a claim or business plan with AI?▼

Provide the claim as an argument and the Skill decomposes it into a testable proposition, then runs parallel research agents for supporting evidence, counter-evidence, base rates, and expert consensus. It finishes with a severity-ranked list of weaknesses and rebuttal material.

How to run a pre-mortem analysis before a decision?▼

The red-team phase enumerates failure modes, hidden assumptions, and the strongest steelmanned opposition argument for your plan. Each anticipated criticism is paired with rebuttal evidence or an honestly stated limitation, so you enter the decision prepared.

Does this Skill require the Parallel Search MCP server?▼

No, Parallel Search MCP is optional. The Skill checks for it first and falls back to WebSearch plus WebFetch if the MCP is not configured, continuing the investigation without stopping.

Why does the Skill call a second LLM through codex?▼

A single model tends to agree with its own conclusions, so the Skill sends the case to codex with an adversarial prompt for an independent critique. Every codex finding is re-verified against primary sources before being accepted or rejected.

What are the limitations of automated evidence gathering?▼

Single-source claims are marked unverified unless corroborated, and search-based evidence can miss paywalled or non-public data. The final report explicitly lists unresolved questions and conditions that could overturn the verdict.