test

Conducts military-grade adversarial analysis by decomposing claims and deploying 32 expert agents to produce steelman arguments and counterarguments.

Security Scan

This instruction-only RedTeam skill is internally consistent with its stated adversarial-analysis purpose, but it asks the agent to read local customization files and make a local HTTP call — users should be aware of those behaviors.

Purpose & Capability

The files and SKILL.md describe a RedTeam/AdversarialValidation workflow (32-agent parallel analysis, adversarial rounds) and the required behaviors match that purpose; there are no unrelated dependencies, credentials, or binaries requested.

Instruction Scope

Runtime instructions explicitly tell the agent to: (1) check for and load user overrides from ~/.claude/skills/CORE/USER/SKILLCUSTOMIZATIONS/RedTeam/ (PREFERENCES.md and resources) and (2) POST a local notification to http://localhost:8888/notify via curl. Reading a per-skill customization directory and emitting a local notification are reasonable for a customizable skill, but they give the skill access to arbitrary files in the specified path and allow it to contact a local service — the user should be comfortable with that.

Install Mechanism

No install spec or code is present (instruction-only). Nothing is downloaded or written to disk by an install step, minimizing install-time risk.

Credentials

The skill declares no required environment variables, credentials, or config paths beyond the optional per-skill customization directory. There are no requests for unrelated secrets or external service keys.

Persistence & Privilege

always:false and no persistent install behaviors are declared. The skill author expects the agent to launch many internal 'agents' and to call other skills (FirstPrinciples, storyexplanation, etc.), which is consistent for a coordination/orchestration skill but increases the amount of downstream tooling and data flows involved — this is normal but expands the effective blast radius.

Assessment

This skill appears to do what it says: coordinate adversarial analyses and syntheses. Before installing, consider: (1) it will look for and load files from ~/.claude/skills/CORE/USER/SKILLCUSTOMIZATIONS/RedTeam/ (so don't store secrets there unless you want them used), (2) it will attempt a local HTTP POST to http://localhost:8888/notify (harmless if you run a notifier, unexpected otherwise), and (3) executing the skill launches many internal agents and may call other installed skills — review downstream skill permissions and any custom files you place in the customization directory. If you want higher assurance, inspect any PREFERENCES.md you put in that folder and ensure no sensitive data is present.

Related: [[AGENTS]], [[skills/pai-redteam/Workflows/AdversarialValidation|AdversarialValidation]], [[skills/pai-redteam/Integration|Integration]]

---

name: RedTeam description: Adversarial analysis with 32 agents. USE WHEN red team, attack idea, counterarguments, critique, stress test. SkillSearch('redteam') for docs.

Customization

Before executing, check for user customizations at: ~/.claude/skills/CORE/USER/SKILLCUSTOMIZATIONS/RedTeam/

If this directory exists, load and apply any PREFERENCES.md, configurations, or resources found there. These override default behavior. If the directory does not exist, proceed with skill defaults.

RedTeam Skill

Military-grade adversarial analysis using parallel agent deployment. Breaks arguments into atomic components, attacks from 32 expert perspectives (engineers, architects, pentesters, interns), synthesizes findings, and produces devastating counter-arguments with steelman representations.

Voice Notification

When executing a workflow, do BOTH:

1. Send voice notification:

   curl -s -X POST http://localhost:8888/notify \
     -H "Content-Type: application/json" \
     -d '{"message": "Running the WORKFLOWNAME workflow from the RedTeam skill"}' \
     > /dev/null 2>&1 &
   

2. Output text notification:

   Running the **WorkflowName** workflow from the **RedTeam** skill...
   

Full documentation: ~/.claude/skills/CORE/SkillNotifications.md

Workflow Routing

Route to the appropriate workflow based on the request.

When executing a workflow, output this notification directly:

Running the **WorkflowName** workflow from the **RedTeam** skill...

Trigger	Workflow
Red team analysis (stress-test existing content)	Workflows/ParallelAnalysis.md
Adversarial validation (produce new content via competition)	Workflows/AdversarialValidation.md

---

Quick Reference

Workflow	Purpose	Output
ParallelAnalysis	Stress-test existing content	Steelman + Counter-argument (8-points each)
AdversarialValidation	Produce new content via competition	Synthesized solution from competing proposals

The Five-Phase Protocol (ParallelAnalysis):

1. Decomposition - Break into 24 atomic claims
2. Parallel Analysis - 32 agents examine strengths AND weaknesses
3. Synthesis - Identify convergent insights
4. Steelman - Strongest version of the argument
5. Counter-Argument - Strongest rebuttal

---

Context Files

• Philosophy.md - Core philosophy, success criteria, agent types
• Integration.md - Skill integration, FirstPrinciples usage, output format

---

Examples

Attack an architecture proposal:

User: "red team this microservices migration plan"
--> Workflows/ParallelAnalysis.md
--> Returns steelman + devastating counter-argument (8 points each)

Devil's advocate on a business decision:

User: "poke holes in my plan to raise prices 20%"
--> Workflows/ParallelAnalysis.md
--> Surfaces the ONE core issue that could collapse the plan

Adversarial validation for content:

User: "battle of bots - which approach is better for this feature?"
--> Workflows/AdversarialValidation.md
--> Synthesizes best solution from competing ideas

---

Last Updated: 2025-12-20