HeathWang

Use when diagnosing bugs, crashes, or unexpected behavior in iOS, React, Python, or Java Spring code. Triggered by error messages, stack traces, crash reports, or when user says "analyze this bug" or "why is this crashing". Performs root cause analysis, evidence gathering, and fix recommendations. Output in Chinese.

Use when the user wants architecture analysis, refactoring opportunities, module-depth review, testability improvements, design debt findings, or durable architecture analytics for a codebase, unless the request or runtime is chat-only, no-file, or read-only.

Use when iOS apps using SwiftGen L10n need localization work - adding keys, checking missing translations, or cleaning unused entries. Triggered by hardcoded strings, missing L10n references, incomplete translations across multiple languages, or unused localization cleanup requests.

Use when the user asks for terse, concise, token-efficient replies while keeping technical accuracy, says "caveman mode", "be brief", "less tokens", "回答短点", "少废话", "减少 token", or invokes /caveman.

Use when debugging hard bugs, intermittent production incidents, customer-reported failures, crashes, broken workflows, thrown errors, unclear regressions, or cases where reproduction or fix confidence is uncertain.

Use when Codex needs to fetch the content of a single Yuque page from a user-provided yuque.com link through a local logged-in Chrome-family browser, especially for private-space documents where token-based API access is unavailable or undesirable.

Use when a user asks to grill, pressure-test, or stress-test a plan against a project's domain language, glossary, CONTEXT.md, ADRs, documented decisions, code behavior, or overloaded terminology.

Use when doing a code review or design analysis from the perspective of Linus Torvalds - applying principles of good taste, simplicity, pragmatism, and zero tolerance for unnecessary complexity. Triggered by "linus review", "brutal review", "is this code good", or requests for opinionated technical critique.

Use when reviewing source files, diffs, commits, pull requests, or bugfixes for correctness issues, regressions, security problems, performance risks, edge cases, or missing tests across any language or framework.

Use when committing git changes, especially when staged files already exist, changes need logical split commits, untracked files need safety checks, commit hooks modify files, or Chinese Conventional Commit messages are required.

Use when the user asks to create a handoff, session summary, context compact, continuation note, 交接文档, 压缩上下文, or instructions for the next agent to resume work later.

Use when converting Java 1.8 classes to Next.js TypeScript interface definitions - handling type mapping, nullability, inheritance, naming conventions, and BaseReq/BaseResponse exclusions.

Use when creating or publishing a PRD from the current conversation context, especially after requirements, implementation decisions, or product scope have already been discussed.

Use when the user wants a plan, design, architecture, or technical proposal stress-tested, pressure-tested, challenged, red-teamed, grilled, 拷问, 方案评审, or asks to "grill me" before implementation.

Use when converting Java 1.8 classes to Swift 5.0+ Codable structs - handling type mapping, nullability, inheritance flattening, BaseReq exclusions, and property naming conventions.

Use when answering technical questions about a codebase - reading and analyzing project files to provide comprehensive, accurate guidance without modifying any files. Triggered by "explain", "how does X work", "why does X happen", or any question about code behavior or architecture.