Redteam — Development agent for Claude Code
Use for anything under redteam/ — building prompt-injection and exfiltration suites, mapping what an injected instruction could make the agent send and where, running adversarial passes against a pinn.
How to install Redteam
Installs to ~/.claude/agents/anapsousa-agent-reliability-lab-redteam.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/anapsousa/agent-reliability-lab/HEAD/.claude/agents/redteam.md -o ~/.claude/agents/anapsousa-agent-reliability-lab-redteam.md Restart Claude Code, or start a new session, for it to be picked up.
What Redteam does
name: redteam description: Use for anything under redteam/ — building prompt-injection and exfiltration suites, mapping what an injected instruction could make the agent send and where, running adversarial passes against a pinned agent, and maintaining the findings log with severity, reproduction and fix. Trigger on "red team this", "injection suite", "can this agent be made to leak", "exfiltration path", "threat model the tools", "log a finding", or any adversarial-security question about t
Alternatives in Development
- Livemsg Gate — セッション間メッセージの主張を裏取りして SEND / HOLD を返す read-only gate 3.1k ★
- Email Wizard — Email campaign orchestration specialist 587 ★
- .github/workflows/redteam.yml — name: Continuous Red Team Assessment on: push: branches: [main, develop] pull_request: branches: [main] schedu 213 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Rfe Scorer
Scores a single RFE issue against quality rubric. Restricted to Read and Write only to prevent prompt injectio
Sonmat Worker
General-purpose worker agent. Discipline is injected via dispatch prompt.
Support
Turns a Sortd user report (email, TestFlight feedback, App Store review) into a reproducible finding for findi
Ct Prompt Craft Auditor
Audits CC instruction files for clarity, actionability, and correct CC behavior — triggers on PROMPT_CRAFT_ADV
Prompt Forager
Context-starved prompt forager for diversity injection. Receives ONE line (a purpose) and deliberately nothing
Typesafe Adversary
Red-teams TypeSafe integrations before users do. Probes the nine documented Jev failure modes, tests prompt in
Related Skills
Reverse Engineer Anything Self Learning Loop
Agent skill: examine a product you're authorised to rebuild, build your own, then test against the original an
Redteam
Adversarial review of the design doc. Run in a fresh session, ideally on a different vendor's model.
Nuguard Redteam
Run authorized non-destructive NuGuard adversarial testing and report evidence.