Results for “red-team”

AI Safety Specialist IA — Expert en sécurité IA (alignment, guardrails, filtrage de contenu, test de biais, red teaming)

Bypasses safety filters on API-served LLMs using jailbreak templates, input obfuscation, and multi-model racing.

Probe RAG applications for prompt injection via poisoned retrieved context and embedding manipulation.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.

Run multi-agent research discourse session. Use when the user requests to "start a research party" or "run a journal club".

Equipment manager for Claude Code projects — inventory local skills/agents/scripts, audit external upstream repos (red-flag + prompt-injection screening), and sync selectively after user approval. Absorbs container-layout skill repos AND single-skill-at-root repos (e.g. a company design-system repo). Project-agnostic.

Run three agents as a newsroom — a Writer, an Editor, and a Fact-checker — that draft, critique, and verify in parallel and argue until the writing survives with zero flags. This is the level above a single self-review loop, for the pieces that matter most. Best run in Claude Cowork against the user's files. Use for high-stakes writing the user wants bulletproof: a newsletter, a launch post, a client email, a public announcement. Trigger whenever the user says 'run the team', 'use the swarm', 'writer editor fact-checker', 'spawn agents to work on this', or wants the strongest possible version of a piece. For a lighter single-agent loop, use red-pen instead.

Never show the user a first draft. Run every writing task through a self-critique loop — draft, attack the draft as the harshest reviewer in the room, rewrite, and repeat until a full review pass finds zero flags — then return only the final plus a change log. Use for any writing the user actually cares about: emails, LinkedIn posts, newsletters, docs, announcements, client messages. Trigger whenever the user says 'run the loop', 'self-critique this', 'make it bulletproof', 'don't give me a first draft', 'be brutal', or hands over a task where quality matters more than speed. This is a single-agent loop; for the three-agent version use the-team.

Makes completion state obvious at the end of every response using a red/yellow/green status block.

Test LLM applications for system prompt leakage using manual payloads, garak, and Promptfoo to extract embedded secrets and routing logic.

Defines the architecture and orchestration primitives for multi-agent teams, covering role allocation, conflict resolution, reward routing, and the operational lifecycle of agent collectives.

N coordinated agents on shared task list using tmux-based orchestration

Use when work requires inter-agent communication, cross-layer coordination, or configurable autonomy with Claude Code Agent Teams

Use when implementing features or fixing bugs with test-driven development. Enforces RED→GREEN→REFACTOR cycle with vertical slicing and multi-agent context isolation. Triggers on: 'implement with TDD', 'write tests first', 'red green refactor', 'test-driven', '/tdd <feature>'. Supports Jest, Vitest, pytest, Go test, cargo test, RSpec, PHPUnit.

Drafts structured goal-loop prompts for long-running agent work with verifiable stop conditions, validation commands, and documentation requirements.

Automate multi-turn adversarial conversations against LLM agents using Microsoft PyRIT, including Crescendo and Tree-of-Attacks-with-Pruning (TAP) attack chains with scorer feedback loops.

Jailbreak API-served LLMs using G0DM0D3 techniques — Parseltongue input obfuscation (33 techniques), GODMODE CLASSIC system prompt templates, ULTRAPLINIAN multi-model racing, encoding escalation, and Hermes-native prefill/system prompt integration. Use when a user wants to bypass safety filters, jailbreak a model, or red-team model robustness.

Jailbreak API-served LLMs using G0DM0D3 techniques — Parseltongue input obfuscation (33 techniques), GODMODE CLASSIC system prompt templates, ULTRAPLINIAN multi-model racing, encoding escalation, and Hermes-native prefill/system prompt integration. Use when a user wants to bypass safety filters, jailbreak a model, or red-team model robustness.

AI/LLM red-team assessment using the OWASP LLM Top 10 (2025) + OWASP AI Testing Guide (AITG v1, Nov 2025) frameworks, plus OWASP MCP Top 10 runtime testing for agentic/MCP targets. Tests prompt injection, jailbreaks, system prompt leakage, sensitive data extraction, excessive agency, improper output handling, model extraction, content bias, evasion, membership inference, MCP token exposure, MCP command injection, and more. Uses four tools in combination: FuzzyAI (single-turn jailbreak fuzzing), PyRIT (multi-turn orchestrated attacks), Garak (probe-based vulnerability scanning), and promptfoo (plugin-based red-team evaluation). Each tool covers different OWASP categories; running them together gives systematic coverage. Includes a conditional MCP reconnaissance phase and a post-access AI infrastructure phase (chained from /post-exploit). Produces: OWASP LLM Top 10 + AITG + MCP coverage matrix, findings per category, architecture diagram of the AI system, PoCs for confirmed exploits. Chains into /gh-export for