outcome-bounded-work

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Outcome-Bounded Work

结果约束型工作

Outcome

目标结果

Shape a live conversation or audit an instruction-bearing artifact so the required result, proof, and safety stay fixed while replaceable implementation choices remain open.
构建实时对话或审核包含指令的工件,确保所需结果、验证依据和安全性保持固定,同时保留可替换的实现选项。

Trigger and boundary

触发条件与适用边界

Use when a conversation, prompt, task, specification, plan, policy, or skill may mix true requirements with one possible way to satisfy them.
Do not use for a simple request with no material method choice. Do not weaken an exact method, explicit user requirement, or higher-priority rule. Use
simplify-skill
when the requested outcome is limited to simplifying a skill package.
Load
starting-point
before the first move through the host's skill-loading capability. It owns the private map of outcome, proof, constraints, starting path, and unknowns, plus the intervention gate for changing routes. This skill turns that map and the source instructions into an inspectable contract without taking over planning or execution.
Use conversation mode for live requests. Use artifact audit mode only when the user asks to inspect an existing instruction-bearing source.
当对话、提示词、任务、规格说明、计划、政策或Skill可能将真实需求与一种可行的满足方式混合在一起时使用。
请勿用于无实质方法选择的简单请求。不得弱化明确的方法、用户明确要求或更高优先级的规则。当请求目标仅为简化Skill包时,请使用
simplify-skill
在首次操作前,通过宿主的Skill加载能力加载
starting-point
。它包含结果、验证依据、约束条件、初始路径和未知项的私有映射,以及用于变更路径的干预闸门。该Skill会将此映射和源指令转换为可审核的合约,而不会接管规划或执行工作。
实时请求使用对话模式。仅当用户要求审核现有包含指令的源文件时,才使用工件审核模式。

Contract fields

合约字段

  • Outcome: The observable end state.
  • Proof: Evidence that decides PASS or BLOCKED.
  • Boundaries: Scope, authority, resources, privacy, cost, and reversibility.
  • Forbidden outcomes: States that must never be accepted.
  • Mandatory methods: Methods fixed by authority, user intent, or the deliverable.
  • Candidate paths: Replaceable ways to pursue the outcome.
  • Adaptation rule: Evidence and approval needed to replace a candidate path.
  • Unknowns: Facts or judgments that can still change the contract.
Use
assets/contract-template.md
when a durable contract artifact is needed. Copy it out of the installed skill before filling it. Do not edit the packaged asset.
  • **目标结果:**可观测的最终状态。
  • **验证依据:**用于判定通过或受阻的证据。
  • **边界条件:**范围、权限、资源、隐私、成本和可逆性。
  • **禁止结果:**绝对不可接受的状态。
  • **强制方法:**由权限、用户意图或交付要求确定的固定方法。
  • **候选路径:**可替换的目标实现方式。
  • **适配规则:**替换候选路径所需的证据和审批流程。
  • **未知项:**仍可能改变合约的事实或判断。
当需要持久化合约工件时,请使用
assets/contract-template.md
。在填写前需从已安装的Skill中复制该模板,不得编辑打包后的资产文件。

Classification rules

分类规则

Classify a source instruction by what would change if it were removed or replaced:
  • It is an outcome invariant when the requested end state would change.
  • It is an evidence obligation when the acceptance decision would lose required proof.
  • It is a boundary when scope, authority, safety, privacy, cost, or reversibility would change.
  • It is a forbidden outcome when it names a state that cannot be accepted.
  • It is a mandatory method when the method is exact, exclusive, contractual, regulatory, reproducibility-related, pedagogical, explicitly required by the user, or itself the deliverable.
  • It is a candidate path when another method could preserve every invariant, proof obligation, boundary, and forbidden outcome.
  • It is unresolved when available evidence cannot distinguish a requirement from a recipe.
Do not demote an imperative instruction on guesswork. Use source wording, authority, context, and retrievable evidence. If the distinction still changes the work materially, ask one concise question.
根据移除或替换源指令后会发生的变化对其进行分类:
  • 若请求的最终状态会改变,则属于结果不变量
  • 若验收判定会失去必要的验证依据,则属于验证义务
  • 若范围、权限、安全性、隐私、成本或可逆性会改变,则属于边界条件
  • 若指定了不可接受的状态,则属于禁止结果
  • 若方法是明确的、排他的、合约规定的、合规要求的、与可复现性相关的、教学性的、用户明确要求的,或是交付物本身,则属于强制方法
  • 若存在其他方法可以保留所有不变量、验证义务、边界条件和禁止结果,则属于候选路径
  • 若现有证据无法区分需求与实现步骤,则属于未决项
不得仅凭猜测降低指令的优先级。需依据原文表述、权限、上下文和可检索的证据。若该区分仍会对工作产生实质性影响,请提出一个简洁的问题。

Conversation mode

对话模式

  1. Build the
    starting-point
    private outcome map.
  2. Infer a provisional contract from the user's words without inventing requirements.
  3. Put the smallest useful contract or worked example in front of the user early.
  4. Label observed facts, reversible assumptions, and open items.
  5. Keep at most one unanswered question live.
  6. Update the same contract from the user's reaction instead of restarting discovery.
  7. Hand off to the matching owner when the contract is accepted or the user asks to act.
  1. 构建
    starting-point
    私有结果映射。
  2. 从用户表述中推导临时合约,不得凭空创造需求。
  3. 尽早向用户展示最精简的可用合约或已完成示例。
  4. 标注观测到的事实、可撤销的假设和未解决事项。
  5. 同时最多保留一个未解答的问题。
  6. 根据用户反馈更新现有合约,而非重新启动需求探索。
  7. 当合约被接受或用户要求执行时,移交至对应领域负责人。

Artifact audit mode

工件审核模式

  1. Read the complete source before proposing changes.
  2. Give every material instruction a stable source ID in source order.
  3. Record its text or locator, authority, class, proposed disposition, and reason.
  4. Preserve invariants, evidence obligations, boundaries, forbidden outcomes, and mandatory methods without loss of meaning or strength.
  5. Keep a recipe mandatory unless evidence supports candidate-path status.
  6. Return the classification ledger and a minimally changed proposed contract.
  7. Do not mutate the source unless the user asks and the rightful editing owner is loaded.
  1. 在提出修改建议前,通读完整的源文件。
  2. 为每个实质性指令按源文件顺序分配一个稳定的源ID。
  3. 记录其文本或定位符、权限、分类、拟处理方式及理由。
  4. 保留不变量、验证义务、边界条件、禁止结果和强制方法,不得弱化其含义或效力。
  5. 除非有证据支持其为候选路径,否则将实现步骤视为强制方法。
  6. 返回分类台账和修改最少的拟议合约。
  7. 除非用户要求且已加载合法编辑负责人,否则不得修改源文件。

Adaptation gate

适配闸门

Replace a candidate path only when:
  • The alternative preserves every accepted contract field.
  • Evidence shows a material benefit or the current path is blocked.
  • The alternative stays within current authority and side-effect limits.
  • Any dropped, delayed, or reprioritized work has user approval.
If these conditions are not proved, keep the current path or mark the route unresolved. A better component is not proof of a better whole outcome.
仅在以下情况下替换候选路径:
  • 替代方案保留所有已接受的合约字段。
  • 有证据表明替代方案具有实质性益处,或当前路径已受阻。
  • 替代方案符合当前权限和副作用限制。
  • 任何取消、延迟或重新排序的工作均已获得用户批准。
若无法满足这些条件,请保留当前路径或标记该路径为未决。单个组件更优并不代表整体结果更优。

Output

输出内容

Show the contract first. In audit mode, follow it with the classification ledger. Keep the difference between the accepted contract, proposed route, and execution evidence visible.
When the next action is text revision, load
meaning-preserving-rewrite
. Route skill-only simplification to
simplify-skill
, plan-only work to
plan
, standing-goal packaging to
goal-prompt
, and execution to the matching domain owner.
先展示合约。在审核模式下,合约后需附上分类台账。需清晰展示已接受合约、拟议路径和执行证据之间的差异。
若下一步操作是文本修订,请加载
meaning-preserving-rewrite
。仅Skill简化工作路由至
simplify-skill
,仅规划工作路由至
plan
,固定目标打包工作路由至
goal-prompt
,执行工作路由至对应领域负责人。

Which worked example should you read?

应阅读哪个示例?

Read
examples/conversation-candidate-path.md
when a live request names an outcome, boundary, and proposed implementation. Read
examples/artifact-audit.md
when a policy or specification may mix fixed approval rules with replaceable tools. Read
examples/simple-request-bypass.md
before adding contract work to a simple request; this is the failure the skill is most likely to cause.
当实时请求指定了结果、边界条件和拟议实现方案时,请阅读
examples/conversation-candidate-path.md
。当政策或规格说明可能将固定审批规则与可替换工具混合在一起时,请阅读
examples/artifact-audit.md
。在为简单请求添加合约工作前,请阅读
examples/simple-request-bypass.md
;这是该Skill最可能引发的错误场景。

Pitfalls

常见误区

  • Treating shorter instructions as automatically better.
  • Preserving every step merely because it was written down.
  • Demoting a safety or evidence rule because it sounds procedural.
  • Replacing a user-selected method without evidence or approval.
  • Inventing proof, constraints, or forbidden outcomes.
  • Treating a contract or proposed rewrite as proof of execution.
  • Editing an audited source before the user authorizes the change.
  • 认为指令越短越好。
  • 仅仅因为步骤被记录就保留所有步骤。
  • 因安全或验证规则听起来像流程性内容而降低其优先级。
  • 在无证据或未获批准的情况下替换用户选择的方法。
  • 凭空创造验证依据、约束条件或禁止结果。
  • 将合约或拟议修订视为已执行的证明。
  • 在用户授权修改前编辑已审核的源文件。

Progressive disclosure

渐进式披露

Read
references/eval-cases.json
before testing, reviewing, or changing runtime behavior. Read
references/generation-contract.md
before changing package structure or native-to-public lineage. Read
evals/cases.json
and the rest of
evals/
only when measuring activation, behavior, failure handling, recovery, or speed. Load
scripts/tests/
before changing validation behavior, then run
mise run ci
; ordinary use does not load
scripts/
.
在测试、评审或修改运行时行为前,请阅读
references/eval-cases.json
。在修改包结构或原生至公开版本的继承关系前,请阅读
references/generation-contract.md
。仅在衡量激活情况、行为表现、故障处理、恢复能力或速度时,才阅读
evals/cases.json
evals/
目录下的其他文件。在修改验证行为前,请加载
scripts/tests/
,然后运行
mise run ci
;日常使用无需加载
scripts/

Verification

验证标准

  • starting-point
    shaped the private outcome map before classification.
  • Every material source instruction has one source ID and a visible class.
  • Preserved requirements retain their meaning, authority, and strength.
  • Every relaxed recipe names the preserved contract and supporting evidence.
  • Unknowns stay visible instead of being resolved by guesswork.
  • The contract can admit a better safe route without allowing a worse result.
  • External actions and source mutations remain behind their normal gates.
  • 分类前已通过
    starting-point
    构建私有结果映射。
  • 每个实质性源指令均有一个源ID和明确的分类。
  • 保留的需求未丢失其含义、权限和效力。
  • 每个放宽的实现步骤均明确了保留的合约和支持证据。
  • 未知项保持可见,未通过猜测解决。
  • 合约可接纳更优的安全路径,同时不会允许更差的结果。
  • 外部操作和源文件修改仍需通过常规闸门审批。