「adversarial」の検索結果

Generate adversarial inputs, edge cases, and boundary test payloads for stress-testing LLM robustness

日本語の概要は準備中です。原文の説明を表示しています。

Multi-agent adversarial code review. Spawns three reviewer subagents (Auditor, Adversary, Pragmatist) on a single model, runs a cross-examination round, then deterministically synthesizes a single ranked report. Use this for non-trivial PRs, security-sensitive changes, refactors that touch many files, or any change where one perspective has obvious blind spots. Trigger phrases include "adverse review", "adversarial review", "multi-perspective review", "review my changes from multiple angles", "panel review". NOT suitable for trivial diffs (typos, dependency bumps, formatting).

日本語の概要は準備中です。原文の説明を表示しています。

N-round opposing-stance debates for trade-off analysis. Assigns pro/con roles to agents, runs structured debate rounds with quality scoring, and produces a moderator synthesis with confidence-rated recommendation. Generalizable to architecture, technology, security, and design decisions.

日本語の概要は準備中です。原文の説明を表示しています。

Force adversarial code review stance that eliminates confirmation bias — reviewer must find issues or re-analyze

日本語の概要は準備中です。原文の説明を表示しています。

Devil's Advocate stress-testing for code, architecture, PRs, and decisions. Surfaces hidden flaws through structured adversarial analysis with metacognitive depth. Use for high-stakes review, stress-testing choices, or when the user wants problems found deliberately. NOT for routine code review. Triggers on "스트레스 테스트", "stress test", "devil's advocate", "반론", "이거 괜찮아", "문제 없을까", "깊은 리뷰", "critical review", "adversarial".

日本語の概要は準備中です。原文の説明を表示しています。

Solve competition math problems (IMO, Putnam, USAMO, AIME) with adversarial verification that catches the errors self-verification misses. Activates when asked to 'solve this IMO problem', 'prove this olympiad inequality', 'verify this competition proof', 'find a counterexample', 'is this proof correct', or for any problem with 'IMO', 'Putnam', 'USAMO', 'olympiad', or 'competition math' in it. Uses pure reasoning (no tools) — then a fresh-context adversarial verifier attacks the proof using specific failure patterns, not generic 'check logic'. Outputs calibrated confidence — will say 'no confident solution' rather than bluff. If LaTeX is available, produces a clean PDF after verification passes.

日本語の概要は準備中です。原文の説明を表示しています。

Use this skill when running or interpreting VCSDD adversarial reviews. Provides calibration guidance for the Adversary agent, finding severity classification, and anti-leniency enforcement patterns.

日本語の概要は準備中です。原文の説明を表示しています。

Use when a plan, feature, or launch needs an explicit adversarial pass to find weak assumptions, abuse cases, or strategic blind spots before execution.

日本語の概要は準備中です。原文の説明を表示しています。

Simulate supply chain and adversarial machine learning attacks by injecting poisoned data or targeted backdoors into training and fine-tuning datasets. Use this skill when assessing the integrity controls of MLOps pipelines or evaluating the resilience of AI models against highly targeted, stealthy manipulation intended to alter model behavior on specific triggers.

日本語の概要は準備中です。原文の説明を表示しています。

Identify and simulate Data Poisoning attacks aimed at degrading or skewing an AI model's accuracy. This skill focuses on Adversarial Machine Learning concepts where attackers inject malicious or mislabelled data points into training or fine-tuning datasets (e.g., feedback loops) to bias the AI.

日本語の概要は準備中です。原文の説明を表示しています。

Run a seven-agent adversarial pipeline against a resume and cover letter before submission. Produces channel-conditional verdicts anchored to the realistic applicant pool, separates materials quality from submission expected value, and ends with a contrarian pass that can overturn the verdict.

日本語の概要は準備中です。原文の説明を表示しています。

Orchestrate a whole-registry sweep of Gaia — fan out 13 parallel audit agents across every skill, run adversarial verification, surface Semantic Fusion candidates, propose new generic skill references, and synthesize a publish-ready HTML report under docs/meta/reports/. Use this skill for broad, systemic analysis across many skills at once: "run a meta sweep", "sweep the meta", "full meta audit", "audit the whole registry against META.md", "widespread nomenclature issues", "find all skills missing a GitHub link", "produce a meta report", "check for evidence type mismatches across the registry", "find skills that document a third-party tool with no attribution", or explicitly types /gaia-meta-sweep. This is the registry-wide macro companion to /gaia-meta-audit (a prioritized queue, single-pass) and /gaia-audit (fixes one skill). Use it when you need the full surface — 13 audit dimensions, adversarial verification, fusion proposals, and a durable findings artifact — not just a queue.

日本語の概要は準備中です。原文の説明を表示しています。

Top-level orchestrator for the Gaia Skill Tree evidence verification pipeline. Run this when you want to do a full evidence pass — collecting raw sources, verifying live GitHub star counts, running Phase 2B benchmark-source verification, adversarially auditing for quality/formatting issues, and checking that every URL is reachable — all in one coordinated sequence. Trigger phrases: "run the evidence pipeline", "full evidence pass", "verify evidence", "evidence verification pipeline", "run ev-pipeline", "check all evidence", "audit the data lake", "evidence quality sweep", "refresh evidence", "validate evidence sources". Also aliased as /evidence-verification-pipeline. Use the individual sub-skills (ev-collection, ev-star-verification, ev-adversarial-audit, ev-link-validation) only when you need to re-run one phase in isolation; for end-to-end work, always start here.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial project planning workflow. Explores the problem space, drafts tickets in batches, then runs sequential adversarial loops — a mandatory UX loop ("should we build this?"), zero or more discretionary specialist loops (security, performance) for projects with architectural implications in those domains, and a mandatory implementer loop ("could we build this?") — to find gaps before implementation. Creates well-specified, batch-tagged tickets upstream only after every applicable loop signs off.

日本語の概要は準備中です。原文の説明を表示しています。

Use when adversarially reviewing a document that makes formal or technical claims — math derivations, physics papers, statistical analyses, benchmark reports, whitepapers. Inventories every equation and quantitative claim, verifies each AS NAMED in the text (never a paraphrase or a neighboring statement), and classifies VERIFIED / REFUTED / UNVERIFIABLE / VACUOUS. Triggers on "check this paper", "verify these claims", "is this derivation right", "review this proof", "audit this benchmark", "does the math hold up". For source-code review see code-review-and-quality; for skill/agent library audits see skill-library-review; for content quality scoring see content-ops.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial implementation review based on Block's g3 dialectical autocoding research. Use when validating implementation completeness against requirements with fresh objectivity.

日本語の概要は準備中です。原文の説明を表示しています。

Applies rigorous adversarial analysis to generate, critique, fix, and consolidate solutions for any problem (technical or non-technical). Use when facing complex problems requiring thorough analysis, multiple solution approaches, and validation of proposed fixes before implementation.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial code review using attacker mindset — trust boundary, attack surface, business logic, and defense evaluation

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial dialectic tool inspired by GAN. Alternates between Generator (creative advocate) and Discriminator (adversarial critic). Supports forced role selection (g/d), intensity modes (hard/soft), and multi-language output. Use for stress-testing ideas, business plans, code, and spec designs.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial code review that breaks the self-review monoculture. Use when you want a genuinely critical review of recent changes, before merging a PR, or when you suspect Claude is being too agreeable about code quality. Forces perspective shifts through hostile reviewer personas that catch blind spots the author's mental model shares with the reviewer.

日本語の概要は準備中です。原文の説明を表示しています。

Use when reviewing PRs/diffs/branches/documents for .NET quality, correctness, performance, security, data access, messaging, and observability. Includes adversarial critical-thinking mode for skeptical/cynical review requests. Keywords: code review, review PR, review diff, critical review, cynical review, adversarial review, production readiness, low GC, AOT, security review.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial critique — devil's advocate, stress-test, honest teardown ('poke holes', 'be brutal', 'was hältst du davon'); explicit request only. Routine code or design review → code-review.

日本語の概要は準備中です。原文の説明を表示しています。

Multi-agent adversarial verification with convergence loop. Two independent review agents must both pass before output ships. Use when output must clear two independent adversarial reviewers before it ships.

日本語の概要は準備中です。原文の説明を表示しています。

Adversarial code review that breaks the self-review monoculture.

日本語の概要は準備中です。原文の説明を表示しています。