AI Agent Skills

Guidelines for writing robust, modern Java tests using JUnit 5, Mockito, and Testcontainers.

Detect and prevent overfitting to historical data via multiple testing corrections and pre-registration. Use when evaluating strategy variants to ensure performance is n…

Prevent train-test contamination, target leakage, and temporal leakage. Use when splitting data, fitting preprocessors, or engineering features for time-series ML.

Test strategy robustness to parameter variation and detect overfitting cliffs. Use when validating that performance is stable across parameter perturbations.

Abuse an LLM agent's tools/functions — coerce it to call tools with attacker-chosen args for SSRF, RCE, data exfil, or privilege abuse. Load when the target is an agent…

Bypass an LLM's safety/guardrails to make it produce restricted output or ignore its policy. Load when testing an AI product's content controls, "jailbreak", "guardrail…

Unbounded-consumption / denial-of-wallet attacks on LLM apps — force runaway tokens, cost, or latency. Load when testing an LLM product's limits/billing, on "LLM DoS", c…

Extract or steal an ML/LLM model's parameters, training data, or system prompt via query access — model stealing, membership inference, training-data extraction. Load wh…

Test gRPC / Protocol Buffers APIs — reflection, method enumeration, authz, and injection. Load on gRPC services (HTTP/2, content-type application/grpc, .proto files, por…

A fast, ordered methodology for assessing an API end to end — REST/GraphQL/gRPC/SOAP — so nothing gets skipped. Load when the target is an API (or recon found one), on "…

面向第三方 AI Agent 的安装与执行指南：如何使用 webmcp-cli 与浏览器页面交互。包含已注入 WebMCP 工具的页面领域专用工具说明。

后室 / 梦核专属空间生成 Skill。用户上传一张人物、角色、宠物或场景图片后，通过 4 个沉浸式选择题提取空间倾向，再结合参考图主体与 Backrooms Level 0 视觉语法，产出可社媒分享的专属后室切片：空间档案卡、AI 配图需求、可选 BGM 方向，并可继续生成真实摄影感后室图片或图生视频。触发词包括：后室测试、我的后室、…

Use when the plan rests on an unproven assumption that could be tested cheaply before committing serious effort — will people turn up, will anyone donate, does this mess…

Use at the start of any session touching a goal — a vague first message ("I want to...", "help me with...", "help me plan...", "what's going on with this"), or any time…

Use before committing to a plan or significant action, when something external changes, or to stress-test reasoning. Red-teams the plan for adversarial and non-adversari…

ALWAYS use when touching ANY DayZ file (.c, config.cpp, mod.cpp, .layout, types.xml) or discussing DayZ modding, Enforce Script, or the Enfusion engine. Activate even if…