'adversarial' 검색 결과

게시물이나 API로 받은 글을 모델에 전달하기 전에 검사합니다. AIDefence로 지시 탈취나 정보 유출을 노리는 입력 등의 위협을 분류하고 판정 신뢰도를 확인합니다.

• 모델 처리 전 사용자 게시물 검사
• API와 Webhook 수신 내용 검사
• 도구 출력에 숨은 지시 검사

Research a question from five search angles, gather sources, and test their claims through multiple verification judgments. The workflow combines overlapping findings and produces a cited report with conclusions ranked by confidence.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Investigating a topic from several perspectives
• Checking claims across multiple web sources
• Preparing a research report with citations

Review code changes or documents through configured review lenses, each representing a different perspective. Findings identify the location, triggering condition, suggested fix, and potential consequence, with reports available as JSON, Markdown, or both.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Reviewing a pull request
• Checking uncommitted changes
• Reviewing specifications or documents

Develop and prioritize early research ideas through independent brainstorming, structured discussion, evidence checks, and critical review. Keep assumptions, uncertainty, dissent, and scoring reasons visible, then record the human decision and appropriate next steps.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Exploring candidate directions for a new research topic
• Organizing ideas before a research team discussion
• Comparing proposals using explicit evaluation criteria

Plan structured adversarial tests for AI applications to assess safety bypasses, data leaks, and unauthorized tool use. Organize test prompts, record findings, and prioritize fixes using reporting guidance and an adaptable testing template.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Assess an AI feature before release
• Compare security behavior of candidate models
• Retest after a prompt injection or jailbreak incident

Coordinate multiple Claude Code workers through an existing Kanban board. This guide describes task dispatch, automated checks before AI review, per-task cost tracking, human intervention, and recovery after restarts; the runtime and integrations require separate installation.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Coordinating multiple Claude Code workers
• Running automated checks before AI code review
• Monitoring task progress and development costs on a board

Check factual claims in AI responses by extracting statements, searching for supporting or conflicting sources, and reviewing common error patterns. Receive a linked report that highlights uncertainties and helps people verify the evidence themselves.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Check statistics and citations in an AI-generated report
• Add ongoing factual checks to a conversation
• Review legal explanations against linked sources

Set up repeatable security tests for LLM apps with Promptfoo and DeepTeam. Connect them to automated build pipelines, detect renewed vulnerabilities to malicious instructions or data leakage, and compare results across model or prompt changes.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Running automated attack tests on every code change
• Blocking merges when security tests fail
• Comparing security across model or prompt versions

Test runnable software through normal and hostile end-to-end scenarios, checking complete behavior across a workflow. Investigate failures, apply fixes, rerun checks, and produce an evidence-backed report covering results, cleanup, and remaining risks.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Check behavior with malformed input
• Verify cancellation and resumption
• Investigate flaky tests

Challenge business plans, prepare for difficult board questions, and work through decisions with painful tradeoffs. This skill identifies assumptions and failure risks, examines stakeholder impacts, and turns postmortems into changes with clear owners and verification dates.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Find weaknesses in a business plan
• Prepare answers to difficult board questions
• Compare options for a difficult management decision

Assess security risks in AI systems by scanning prompts for known attack patterns and scoring data exposure, poisoning, and tool abuse risks. Map findings to MITRE ATLAS, an AI threat framework, and identify protective controls.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Scan application prompts for injection and jailbreak patterns
• Review malicious instructions in retrieved documents
• Assess training data exposure and poisoning risks

Review agent-written code changes before merge, focusing on concrete correctness defects. Uses independent subagent reviews, targeted verification, and fresh checks after substantial repairs, with a deeper evidence-based process for security, concurrency, persistence, or data-loss risks.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Reviewing agent-written changes before merge
• Rechecking a substantial bug fix
• Examining security, concurrency, or data-loss risks

Retune a collection of skills for a new model using archived runs and controlled comparisons. Establish normal measurement variation, register success criteria before editing, and make targeted changes until measured behavior meets the criteria or a specific limitation is reported.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Investigating skill regressions after switching models
• Measuring the effect of instruction changes
• Separating behavior improvements from normal run variation

Delegate coding, analysis, refactoring, or reviews from Claude Code to Codex CLI. The skill explains unattended execution, permission settings, output capture, parallel tasks, and independent checks of changed files and test results.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Delegate implementation while retaining planning responsibility
• Get a second opinion on code changes
• Split independent tasks across separate git worktrees

Delegate coding, analysis, planning, and reviews from Codex to Claude Code through its command-line interface. Capture responses, resume sessions, or run independent tasks in parallel, then verify changed files and test results independently.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Delegating implementation or refactoring from Codex
• Getting another model's review of code changes
• Continuing delegated work across multiple sessions

Compare options through a debate among agents given distinct expertise and positions for the topic. Choose a three-round discussion with a visible transcript and synthesis, or a one-round check that presents initial perspectives.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Compare technical options from several perspectives
• Review an API design with a quick perspective check
• Examine an important decision through structured debate

Write and review launch prompts for agents working autonomously for hours or days. Define successful deliverables, excluded partial results, stopping rules, worker handoffs, and evidence-based review requirements so completion can be assessed against explicit criteria.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Turning vague requests into precise task briefs
• Defining stopping rules for long autonomous runs
• Specifying deliverables for parallel workers

Build evaluation suites for AI features with the Axiom AI SDK. Generate test files, scoring functions, test cases, and configurable parameters from descriptions and existing code to compare changes and monitor output quality in production.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Add regression evaluations for an AI feature
• Score classification, retrieval, and structured outputs
• Compare model and parameter configurations

Review a pull request or code diff from several angles, then have separate agents test each potential finding. Receive a ranked findings table, confidence assessments, a merge recommendation, and concrete follow-up work.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Review a pull request before merging
• Check changed code for lost behavior or broken callers
• Assess duplicate code, complexity, and wasted work

Run two independent reviewers against the same fixed target, then compare their findings. With approval, address severe issues confirmed by both and reassess the fixes, using at most two correction rounds before a final verdict.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Request two independent reviews of a specific target
• Separate confirmed issues from disputed findings
• Fix severe issues confirmed by both reviewers

Validate security patches by comparing vulnerable and patched code against exploit tests, related attack variants, and normal behavior. Produce reproducible logs showing remaining flaws, regressions, newly introduced security failures, and gaps that still need testing.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Validate an AI-generated security patch before review
• Check related attack variants for incomplete fixes
• Check whether a fix breaks normal behavior

Review local branches, pull requests, commits, or files and address issues. The skill selects a single-agent or multi-agent workflow based on the environment, with automatic fixes for lower-risk issues and confirmation for high-risk changes.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Reviewing changes on a local branch
• Starting a pull request review from its number or URL
• Checking a specific commit or file

Create a multi-panel research figure from a clear claim and fixed data references, or start from an existing figure. Plan the layout, coordinate panel creation, inspect the composite, and revise affected panels through independent review.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Designing a multi-panel figure around a research claim
• Reworking the layout of an existing figure
• Checking readability after combining panels

Assess the storage and retrieval layers of RAG systems, which search documents to support AI answers. Test persistent document poisoning, cross-tenant access, source-text leaks, and manipulated search rankings using independently verifiable evidence.

한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.

• Reviewing shared knowledge bases and document chatbots
• Testing access isolation between accounts or organizations
• Checking source text and metadata in search results