agent-roles-spec

Fix bugs by reproducing the failure first, making the smallest correction, and proving the regression is covered. Use for bug reports, failing tests, and regressions.

Fix failing tests or CI by reading logs, identifying the smallest failing surface, making a minimal change, and rerunning targeted verification. Use when given CI logs, failing commands, or broken checks.

Scan relevant source, tests, commands, and existing project patterns before editing code. Use before non-trivial implementation, bug fixing, or refactoring when the right local pattern is unclear.

Control error handling, defaults, retries, compatibility paths, and fallback behavior. Use before adding catch blocks, defaults, shims, optional values, retries, or graceful degradation.

Prevent large-file growth and mixed-responsibility edits. Use when adding significant logic to a large file, mixed-purpose module, or file with repeated conditionals.

Implement the smallest safe repo-native code change for the requested behavior. Use for feature work, small refactors, and multi-file implementation tasks.

Follow existing repository style, commands, naming, file layout, test patterns, dependency conventions, and code shape. Use when implementing in an unfamiliar repo or module.

Gate dependency additions, removals, upgrades, and lockfile changes. Use before changing dependencies, package managers, SDKs, generated lockfiles, or install commands.

Check current official documentation before writing framework, library, CLI, or API code where version-specific correctness matters. Use when APIs are unfamiliar, recently changed, or user asks for current best practice.

Add or update behavior-focused tests for code changes and run appropriate verification. Use when behavior changes, logic is added, or existing behavior could regress.

Create or update project documentation according to CCB document routing rules. Use when Claude asks to write docs, route documents by the docs structure contract, or archive implementation details.

Execute CCB implementation, explore, or consult tasks from Claude based on spec/docs, validate the result, and return a compact receipt. Use when Claude sends a CCB task with execute/explore/consult mode.

重新分析 Requirement 原文和已生成人读设计/任务文档，修正理解漂移。

记录用户对某个审查门、拆分草案或风险事项的明确批准。

在 review 通过后固化完成证据、风险和后续建议。

建立 autonomous-batch 授权范围和停止边界的指令入口。

取消 Requirement 或子任务的大状态指令入口。

暂缓 Requirement 或子任务的大状态指令入口。

把已确认的子任务派给执行 agent 的指令入口。

CCB 主流程入口：按用户意图在 7 个节点中选择工作模式，推进需求分析、设计、拆分或后续执行。

初始化项目级 CCB plugin 工作区，使 plugin 可脱离 Console 独立运行。

将用户已审查通过的 Requirement breakdown draft 物化为子任务。

/ccb:su-flow 的 deprecated alias；保留到 v1.5 grace window。

对低风险已通过审查事项执行快速归档，但仍保留证据和审计。

重新激活 cancelled/deferred 主体的大状态指令入口。

AI-native 自检 docs 人读真相源、.ccb 协调件与 Console 投影漂移，并按审批结果执行修复。

从 docs 人读真相源与 .ccb 协调件恢复 paused/deferred/in-progress 主体的上下文。

审查执行回执、diff 和验证证据，决定通过、返工、replan 或升级用户。

根据用户反馈重写 Requirement breakdown draft。

从 docs 人读真相源与 .ccb 协调件读取当前项目、Requirement 或子任务状态。

Plan, install, boot, control, inspect, reset, and clean up Android emulators, Android Virtual Devices, iOS Simulators, Expo Orbit, React Native, Expo, and Flutter virtual-device workflows with explicit host approval.

Manage and diagnose the CCB adapter's optional npm Archi CLI, bundled Hippo/llmgateway capabilities, and role tool readiness.

Produce architecture improvement advice from architecture evidence and source inspection.

Review current changes for architecture risk, boundary pressure, coupling, duplication, and maintainability impact.

Select architecture and code-review evidence sources before an archi review. Use when the user asks which tools to use, wants less dependence on Architec, needs architecture/code review readiness, or needs a review plan across direct source reading, project-native checks, dependency analysis, SAST, and architecture fitness tools.

Run or read full-project architecture analysis and summarize structural weaknesses, hotspots, and maintainability risks.

Produce goal-driven architecture guidance for a specific boundary, refactor, subsystem, or maintainability objective.

Conducts multi-axis code review. Use before merging any change. Use when reviewing code written by yourself, another agent, or a human. Use when you need to assess code quality across multiple dimensions before it enters the main branch.

Find deepening opportunities in a codebase, informed by the domain language in CONTEXT.md and the decisions in docs/adr/. Use when the user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more testable and AI-navigable.

Use when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation