adversarial-testing

Plugin adversarial-testing from marketplace adversarial-testing-tools in axellab/adversarial-testing-toolkit. Commands read from the repo's manifest — 1 month ago.

Skills (6)

• adversarial-mode

  Toggle persistent Adversarial Testing mode on or off for this project. When ON, every assistant response gains a structured "how this can fail" red-team section automatically (enforced by a deterministic hook). Use when the user says things like "adversarial mode on/off", "activá el modo adversarial

• deep-audit

  Run a comprehensive, multi-agent adversarial audit of a target (feature, module, service, or change) across all dimensions in parallel — functional weak points, security, resilience/performance, and LLM robustness — then adversarially verify each finding to drop false positives before synthesizing o

• generate-tests

  Generate tests for a target (file, feature, or change), matching the project's own framework and conventions, then run them and report results with evidence. Use when the user says "write tests for X", "generá y corré los tests", "add test coverage", or wants actual executable tests — not just recom

• redteam

  Run a focused adversarial critique of one artifact — a plan, spec, code file/change, or LLM prompt — and get back only the structured "how it can fail" analysis. Use when the user says "red-team this", "attacá este plan/código/prompt", "how could this break?", "poke holes in this", or wants an adver

• spec-attack

  Adversarially review spec-driven-development artifacts before they turn into code — attacks a spec, plan, or task list (GitHub Spec Kit spec.md / plan.md / tasks.md, or any spec/plan document) for ambiguity, unstated assumptions, missing and conflicting requirements, untestable acceptance criteria,

• test-plan

  Analyze code (a file, feature, module, or change) and produce a prioritized, risk-ranked plan of the tests worth writing — the weak points, critical paths, and recommended cases by type (unit/integration/e2e/property). Use when the user asks "what should we test?", "qué pruebas recomendás", "where a

Subagents (6)

• llm-robustness-auditor

  Adversarial auditor for LLM-powered features, prompts, and agents. Probes prompt injection, jailbreaks and guardrail bypass, instruction-hierarchy violations (data treated as commands), data exfiltration (system prompt/secrets/other users' context), tool and action abuse, hallucination/grounding gap

• redteam-critic

  Adversarial critic that red-teams any single artifact — an implementation plan, a spec, a code change/file, or an LLM prompt — and returns ONLY the structured failure analysis (how it can break), never a rewrite or the primary answer. Use it whenever you want a focused "how could this fail?" pass ov

• resilience-auditor

  Resilience, reliability, and performance auditor. Probes how the system behaves under stress and failure — dependency outages, timeouts and retry storms, load and scaling limits (N+1, unbounded memory, pool exhaustion), resource leaks, partial-failure and consistency gaps, backpressure, large data v

• security-auditor

  Security-focused adversarial auditor. Examines code, configuration, and design for the attack surface — injection, XSS/CSRF, broken authN/authZ (including IDOR and path traversal), secrets and crypto handling, SSRF and unsafe deserialization, sensitive-data exposure, dependency and config risks, and

• test-engineer

  Detects the project's stack and test framework, generates tests that match the repo's existing conventions, runs them, and reports results with evidence. Give it a target (file/feature/change) or a test plan from weakpoint-scout and it turns weak points into real, executable tests. Use when the user

• weakpoint-scout

  Functional weak-point analyst. Reads code (a file, module, change, or feature) and maps where it is most likely to break — edge cases, invalid states, fragile flows, unhandled errors, concurrency and idempotency gaps — then recommends the highest-value tests to write. Analysis and recommendations on

Hooks (2)

• SessionStart

  powershell -NoProfile -ExecutionPolicy Bypass -File "${CLAUDE_PLUGIN_ROOT}/scripts/locate-core.ps1"

• UserPromptSubmit

  powershell -NoProfile -ExecutionPolicy Bypass -File "${CLAUDE_PLUGIN_ROOT}/scripts/adversarial-mode.ps1"

Similar plugins