Machine learning

Use when integrating ONNX/Sentis/Unity Inference in Unity: model assets, IWorker, runtime performance, or com.unity.sentis packages.

When asked to estimate something unstated, produce a legible correctable evidence chain rather than refusing or fabricating a number.

Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".

定位并操作当前 LLM Wiki 学习知识库。用户要求查询、炼化、收录、整理或体检知识库时使用；先定位仓库根目录并阅读 AGENTS.md。

Use to audit a Unity project's AI/ML stack: packages, models, scripts, configs, and release risks.

Use to find memory/disk pressure from assets: large files, textures, audio, video, ML models, and Addressables that should be remote.

Use for on-device or server inference that is not tied to one vendor model: ONNX Runtime, Sentis, Barracuda legacy, custom .onnx/.tflite/.pb assets, and…

Use for NavMesh, AI Navigation package, NavMeshAgent, NPC pathfinding, off-mesh links, or gameplay AI movement (non-ML).

Add an AI agent / chatbot / RAG backend on Convex with @convex-dev/agent (durable threads, messages, tools, vector search).

Use when a CTF challenge centers on AI or LLM security rather than a conventional web, crypto, or reverse-only path.

Perform CWE-grounded security code reviews and precise weakness mapping using a locally derived MITRE CWE corpus, relationship graphs, mapping notes…

Use LM Studio as a local LLM backend with Pydantic AI for structured output extraction.

Find an existing Expanso job or pipeline recipe before writing one from scratch, and report honestly how far it has been proven.

Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".

Developer reference for connecting to OpenAI, Anthropic, Gemini, DeepSeek, GLM, MiniMax, MiMo, Mistral, xAI, Qwen, Kimi, Cohere, OpenRouter, Together, and…

Run a read-only, evidence-based audit of an LLM or agent application for prompt injection and the failures that make injection matter.

Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".

Super caching: domain router for web (browser, CDN, service worker, HTTP Cache-Control, BFF/API route cache), mobile (on-device, offline, image cache…

Turns a compute run's energy evidence into a typed receipt (MEASURED from counter samples, REPORTED from vendor totals or TDP estimates, UNAVAILABLE when…

Scores binary predictions (Brier, log loss, ECE, MCE, AUROC, accuracy, reliability bins, per-cohort) or saved categorical outputs joined to retained targets…

Use the steipete/summarize CLI to summarize URLs, local files, stdin, YouTube links, podcasts, and media with LLM models.

Builds generative AI applications on Amazon Bedrock.

Rendered guardrail framework: keep a repo's hard rules in one-rule-per-file GRL sources behind a compact router, and render only the rules a task needs.

Use for "interrogate", "adversarial review", "multi-model review", "challenge this", "stress test this code", "find blind spots", or "tear this apart".