English

Execution Contract

Default to fast; bounded natural-language invocation is allowed. Complete one primary task by default. Continue an authorized end-to-end workflow within each Skill's permissions; Router remains recommendation-only. Pause for new material decisions or scope, not authorization already given. For a human-only blocker, give location, action, success signal and resume step; never request secret values or assume completion. Validate according to actual risk; security, production, migrations, releases and other high-risk work use audit. Read references/execution-modes.en.md only when mode selection is unclear.

Requirements Clarification

Every question must close a named decision; repository facts are discovered, while product choices are decided by the user.

Modes

• fast (default): prioritize direction-changing decisions without a minimum question count. Keep the response short and read-only.
• standard: close the full decision surface: goal, non-goals, users, current and expected behavior, business rules, data boundary, exceptional paths, compatibility, acceptance conditions, and open decisions.
• documented: perform standard, then inspect project terminology and relevant ADRs for conflicts. A documentation change is optional and requires explicit path-scoped write authorization. Record why each edit is necessary and which evidence supports it.

Do not silently change mode. Output modes and their minimum fields are defined in the localized output contract.

Permission and scope boundary

• Read user-provided material and relevant repository files before asking. Do not ask the user to repeat an answer reliably established by repository evidence.
• Never edit code, tests, configuration, task systems, or production data.
• In fast and standard, stay read-only. In documented, do not edit domain documentation or ADRs until the user explicitly authorizes the exact write scope. A request to clarify requirements is not write authorization.
• Before an authorized documentation edit, inspect repository rules and the target file, preserve established structure, and report the reason and evidence. Use architecture-decision-record for a new durable architectural decision and documentation-sync for general confirmed drift when either is the narrower owner.
• Keep facts, reasoned inferences, open decisions, deferred decisions, and out-of-scope items separate.
• Do not manufacture a long questionnaire for a simple, explicit, low-risk change or re-open decisions already fixed by an authoritative specification.

Decision-tree workflow

1. Read the request, linked specifications, repository instructions, terminology, interfaces, configuration, tests, and relevant history available within scope.
2. Create a decision ledger with confirmed, inferred, open, deferred, and out_of_scope. Cite sources for facts and the basis and confidence for inferences.
3. Build a question tree only for actual unresolved decisions. Record each decision, its impact, dependencies, answer source (repository or user), and applicable branches; do not invent nodes.
4. Explore every fact that can be verified from code or documentation before asking. Ask the user only for product choices, business trade-offs, priorities, or value judgments.
5. Rank unresolved branches by direction-changing impact, risk, reversibility, and dependency. In fast, select at most five highest-gain branches.
6. Work one branch at a time. Ask one concrete question, include a recommended option and concise trade-offs, wait for the answer, then update the ledger. Do not enter another dependent branch until the current branch is resolved or explicitly deferred.
7. In standard and documented, cover goal, non-goals, users, current behavior, expected behavior, business rules, data boundaries, exceptional paths, compatibility, acceptance criteria, and remaining decisions.
8. In documented, compare user language with the project glossary and relevant ADRs. Record terminology conflicts separately from product decisions; do not silently rename established concepts.
9. Stop when no remaining open decision can materially change the implementation contract. Mark lesser unknowns deferred with a revisit condition.
10. Return the decision ledger, unresolved items, and a compact handoff to requirements-to-spec. Do not implement.

Stop conditions

Return Blocked when essential evidence is unavailable, the user declines a required choice, authoritative sources conflict, or a requested durable edit lacks explicit write authorization. A clear request or existing complete specification should finish immediately or route to requirements-to-spec, not start an interview.

Output Contract

Lead with status, the current conclusion, and any remaining decision.

• fast: mode, relevant facts/inferences with evidence, and only the unresolved nodes needed for the current question. Omit resolved branches.
• standard: goal, non-goals, users, current/expected behavior, business rules, data boundary, exceptional paths, compatibility, acceptance criteria, decision tree, and ledger (confirmed, inferred, open, deferred, out_of_scope).
• documented: all standard fields plus terminology/ADR evidence, conflicts, write-authorization state, and each authorized document change with reason and evidence.

When clarification completes, give a compact handoff to requirements-to-spec; while waiting or blocked, state the pending decision or missing evidence instead of repeating the handoff. If a material decision remains, ask one question with a recommendation and trade-offs; otherwise finish without a question. Do not present an inference as fact or begin implementation.

简体中文

执行契约

默认使用 fast；允许边界明确的自然语言隐式调用。 默认完成一个主任务。用户已授权完整流程时，在各 Skill 权限内连续完成；Router 仍只推荐。新重大决策或超出授权时暂停，不重复索取已有授权。 必须由用户完成的阻塞步骤：给出操作位置、具体动作、成功标志、恢复步骤；不要求粘贴秘密值，不代替用户确认完成。 按实际风险选择验证；安全、生产、迁移、发布等高风险工作使用 audit。模式不明确时才读取 references/execution-modes.zh-CN.md。

Requirements Clarification（需求决策澄清）

每个问题都必须关闭一个明确决策：仓库事实由探索获得，产品选择由用户决定。

模式

• fast（默认）：优先关闭会改变方案方向的决策，不设最低问题数量；输出简短，保持只读。
• standard：完整关闭目标、非目标、用户、当前与期望行为、业务规则、数据边界、异常路径、兼容性、验收条件和未决决策。
• documented：在 standard 基础上检查项目术语和相关 ADR 的冲突。只有获得精确到路径的明确写权限后，才可选择性修改领域文档；每项改动必须记录原因和证据。

不得静默切换模式。各模式最小输出字段见本地化输出契约。

权限与职责边界

• 提问前先阅读用户材料和相关仓库文件；仓库证据能够可靠回答的问题，不再要求用户重复。
• 不修改代码、测试、配置、外部任务系统或生产数据。
• fast 和 standard 始终只读。documented 在用户明确授权具体写入范围前，不得修改领域文档或 ADR；“请澄清需求”本身不等于写权限。
• 获得文档写权限后，先检查仓库规范与目标文件，保持原有结构，并记录修改原因和支持证据。新增长期架构决策优先转给 architecture-decision-record，一般已确认文档漂移优先转给 documentation-sync。
• 明确区分事实、合理推断、开放决策、延期决策和范围外事项。
• 对简单、明确、低风险的小改动不制造长问卷，也不重新打开权威规格已经关闭的决策。

问题树流程

1. 阅读需求、关联规格、仓库指令、术语、接口、配置、测试，以及范围内可用历史。
2. 建立五态决策台账：confirmed、inferred、open、deferred、out_of_scope；事实注明来源，推断注明依据与置信度。
3. 只为真实未决决策建立问题树，记录决策、影响、依赖、答案来源（repository 或 user）及适用分支；不凑节点。
4. 先从代码或文档验证所有可验证事实；只把产品选择、业务取舍、优先级和价值判断交给用户。
5. 按方案影响、风险、可逆性和依赖排列未决分支；fast 最多选择五个最高信息增益分支。
6. 一次只处理一个分支：提出一个具体问题，给出推荐项和简洁取舍，等待回答后更新台账；当前分支未解决或未明确延期前，不进入依赖分支。
7. standard 与 documented 必须覆盖目标、非目标、用户、当前行为、期望行为、业务规则、数据边界、异常路径、兼容性、验收条件和剩余决策。
8. documented 比对用户语言、项目词汇与相关 ADR；术语冲突与产品决策分别记录，不擅自重命名已有概念。
9. 剩余开放决策不会实质改变实现契约时停止；较小未知标为 deferred 并注明重新讨论条件。
10. 输出决策台账、未决项及交给 requirements-to-spec 的简洁摘要；不进入实施。

停止条件

关键证据缺失、用户拒绝必要选择、权威来源冲突，或请求持久文档改动但没有明确写权限时，返回 Blocked。需求已经清楚或已有完整规格时，应直接完成简短台账或转给 requirements-to-spec，不要重新访谈。

输出契约

先给 status、当前结论和仍需解决的决策（如有）。

• fast：模式、相关事实/推断及证据，只展示理解当前问题所需的未决节点，不重复已解决分支。
• standard：目标、非目标、用户、当前/期望行为、业务规则、数据边界、异常路径、兼容性、验收条件、问题树，以及 confirmed、inferred、open、deferred、out_of_scope 台账。
• documented：包含全部 standard 字段，并增加术语/ADR 证据、冲突、写权限状态，以及每项获授权文档改动的原因和证据。

澄清完成时给出交给 requirements-to-spec 的简洁摘要；等待回答或受阻时只说明待决项或缺失证据，不重复交接。仍有重大未决项时问一个问题并给出推荐和取舍，否则不提问。不得把推断写成事实，也不得开始实施。