## 摘要 / Summary
补问、原话解释与真人确认分离；增加版本能力预检、sourceField/evidenceLinks、有限语义用例及正确工件依赖顺序。
Separate answers, interpretation and human confirmation; add capability
discovery, explicit source/evidence bindings, finite semantic checks and
artifact ordering.
发现描述覆盖“仅澄清”入口；有歧义的原话保持缺口，不把“不确定”塞入精确枚举或规范化值。按当前对话语言加载框架与固定事实门尾注。
当前请求中已唯一确定的值会写成类型化、待真人确认的 `node.value`；只有未知或仍有歧义的事实才保持 `needs_input`。
Discovery covers clarify-only requests. Ambiguous raw answers remain
unresolved, rather than becoming exact placeholder values. The framework
and fixed fact-gate footer follow the current conversation language.
Exact values uniquely supplied by the current request become typed,
human-reviewable `node.value` proposals; only genuinely unknown or
ambiguous facts remain `needs_input`.
应用/服务重启或重试 fork 后，入口必须在当前父轮重新加载 Skill，并在委派前用 ToolCatalog 核实阶段工具；子 Agent
不得突破未激活父级的能力上限。
After an app/service restart or retry fork, the entry Agent reloads the
Skill in the current parent turn and verifies stage tools before
delegation; a child never expands an unactivated parent ceiling.
新的 decision-grade `OptimizationSpec` 强制使用 v2，并完整声明 `semantic_contract`
中的 solve intent 及所有 objective/variable/constraint/data reference
的单位/维度；legacy v1 仅只读兼容。
New decision-grade `OptimizationSpec` artifacts use v2 and a complete
semantic contract for solve intent plus every material objective,
variable, constraint and data reference; legacy v1 remains read-only
compatibility.
## 验证 / Validation
- Scoped Skill/i18n/catalog check passes with zero errors. Existing
repository warnings remain.
- Publication safety check passed, including hidden working-tree files,
new paths, branch/commit metadata and collaboration text.
- 双语说明与 source hash 已同步，版本更新为 2.7.0；缺少新平台契约时保持普通澄清，不模拟缺失门禁。
- Bilingual bodies and source hashes are synchronized at version 2.7.0.
Older clients stay in plain-text clarification when capability discovery
fails.
## 状态 / Status
Runtime acceptance is still in progress; draft only. No customer data,
credentials, solver software or deployment settings are included.
真机验收仍在进行，本 PR 暂为草稿。保留外部求解器许可证、部署和费用披露；本变更不包含求解器、客户数据或凭据。
---------
Co-authored-by: yige <yige@yigedeMacBook-Neo.local>
Use this Skill when a user naturally asks about workforce efficiency, staffing, shifts, service coverage, hierarchical resource allocation, performance targets, centralized task scheduling, operations research, or general LP/MILP. Govern the request through DecisionWorkspace, versioned multi-agent artifacts, guarded compile/solve, independent validation, and human approval. MindOptSolve is only a compatible connector Tool: the MindOpt solver software, applicable license, separate deployment, and related fees are external and not bundled. 用户自然提到人效、人员配置、排班、服务范围、资源划分、绩效目标、任务调度、运筹优化或 LP/MILP 时使用；通过 DecisionWorkspace、人机确认、版本化制品、受保护求解、独立验收和真人批准完成治理；MindOpt 软件、适用许可证、独立部署及费用不随本技能或客户端提供。
排班、工时上限、人员配置的第一轮澄清（含“只澄清、暂不求解”）也先用 Skill 工具加载本技能，再用 DecisionWorkspace 补问和真人确认卡；普通 AskUserQuestion 不能代替业务事实闭环。Load this Skill for the first clarification of staffing, shifts, work-hour limits, service coverage, resource allocation, targets, task scheduling or LP/MILP, including clarify only / do not solve yet requests. Use DecisionWorkspace question and human-confirmation cards for model-changing facts; generic AskUserQuestion is not a substitute. Clarification needs no connected solver. Actual MindOpt solving requires a separately licensed/deployed solver and configured connector; software, licenses, hosting and fees are not included. MindOpt 软件、许可证、部署及相关费用需使用方另行取得或承担。
compatibility: >-
compatibility: >-
Requires a separately installed or deployed MindOpt solver, a valid license obtained under the official terms, and configured solver.mindopt connections; commercial licenses and operating costs are purchased separately when applicable. 需要外部安装或部署 MindOpt、按官方条款取得有效许可证并配置 solver.mindopt 连接；适用的商业许可及运行费用需另行采购承担。
Requirement clarification works without a connected solver. Actual MindOpt solving requires separately installed or deployed MindOpt, a valid license under its official terms, and configured solver.mindopt connections; commercial licenses and operating costs are separate when applicable. 需求澄清不要求先连接求解器；实际 MindOpt 求解仍需外部安装部署、有效许可证及 solver.mindopt 连接，适用的商业许可和运行费用另行承担。
short_desc: Co-manage workforce models, guarded solving, and independent validation; MindOpt is externally licensed
short_desc: Co-manage workforce models, guarded solving, and independent validation; MindOpt is externally licensed
description: >-
description: >-
When users naturally request service coverage, hierarchical resource allocation, performance targets, centralized task scheduling, staffing, shift planning, or general LP/MILP, use DecisionWorkspace humanconfirmation, exact-version artifacts, entry Agent binding, peer review, execution guards, independent validation, and human approval to create a recoverable governance chain. When MindOpt is selected, actual solving still requires the user to obtain an applicable license, deploy MindOpt, and configure its connector; the client and this Skill do not include the solver, license, hosted compute, or related fees.
Load this Skill for the first clarification of staffing, shifts, work-hour limits, service coverage, resource allocation, targets, task scheduling or LP/MILP, including clarify only / do not solve yet requests. Use DecisionWorkspace question and human-confirmation cards for model-changing facts; generic AskUserQuestion is not a substitute. Clarification needs no connected solver. Actual MindOpt solving requires a separately licensed/deployed solver and configured connector; software, licenses, hosting and fees are not included.
@@ -74,7 +74,8 @@ Turn natural-language workforce-efficiency requests into reviewable and recovera
- The platform retains direct personal `MindOptSolve` and compile-option overrides only as a backward-compatible, non-decision-grade expert path outside this Skill's governed team workflow. Never present an unbound personal call as reviewed, human-confirmed, recoverable, or decision-grade; team runs must use the guarded `OptimizationSolve` path and solver limits recorded in the committed specification.
- The platform retains direct personal `MindOptSolve` and compile-option overrides only as a backward-compatible, non-decision-grade expert path outside this Skill's governed team workflow. Never present an unbound personal call as reviewed, human-confirmed, recoverable, or decision-grade; team runs must use the guarded `OptimizationSolve` path and solver limits recorded in the committed specification.
- Before promising or requesting an actual solve, verify that the external connector is configured and ready, required capabilities are available, and the deployment has a valid applicable license. A registered Tool name alone is not evidence that the solver is installed, licensed, reachable, or paid for.
- Before promising or requesting an actual solve, verify that the external connector is configured and ready, required capabilities are available, and the deployment has a valid applicable license. A registered Tool name alone is not evidence that the solver is installed, licensed, reachable, or paid for.
- If the external dependency is unavailable, state which prerequisite is missing and stop before the solver call. You may still finish requirement clarification and produce reviewable `SceneSpec`, `DataContract`, and `OptimizationSpec` artifacts for later execution, but must not fabricate a `SolveResult`, feasibility, optimality, or benefit claim.
- If the external dependency is unavailable, state which prerequisite is missing and stop before the solver call. You may still finish requirement clarification and produce reviewable `SceneSpec`, `DataContract`, and `OptimizationSpec` artifacts for later execution, but must not fabricate a `SolveResult`, feasibility, optimality, or benefit claim.
- Before routing or modeling, the natural-language entry Agent must read the [requirement-clarification framework](references/requirement-clarification-framework.md) in full and follow its real-decision, mandatory-question, and conditional-question branches.
- Before routing or modeling, read the framework matching the user's current conversation language in full: for Chinese use [中文澄清框架](references/requirement-clarification-framework.zh-CN.md), and for English use the [English framework](references/requirement-clarification-framework.md). Follow its real-decision, mandatory-question and conditional-question branches. The Skill's default metadata locale does not override the user's language; use that framework's matching-language fixed footer.
- Skill activation is intentionally scoped to the current service-process lifetime and active parent turn. After an app/service restart or retry fork, the current entry Agent must load this Skill again before checking hidden optimization tools or delegating a stage owner; verify `TeamArtifact` and the stage-required optimization tools in `ToolCatalog` before delegation. A child Agent loading the Skill cannot expand a ceiling inherited from an unactivated parent. Do not persist activation, widen default tools, switch source/scope to evade the ceiling, or ask a child to fabricate artifacts when capability is absent.
- First read the AgentFS user profile, preferences, and relationship memories already injected into the current context. Choose professional, business-guided, or evidence-insufficient adaptive language only from user-confirmed, current, non-conflicting evidence about expertise or communication preference. Employer, job title, one use of jargon, or model inference is not sufficient evidence.
- First read the AgentFS user profile, preferences, and relationship memories already injected into the current context. Choose professional, business-guided, or evidence-insufficient adaptive language only from user-confirmed, current, non-conflicting evidence about expertise or communication preference. Employer, job title, one use of jargon, or model inference is not sufficient evidence.
- Professional language may expose the complete structured information contract at once and accept a batch answer. Business-guided language uses plain-language groups in impact order for as many turns as needed. When evidence is insufficient, show a neutral coverage outline and ask the user's preference. Every mode maintains the same complete question map; never omit a model-changing item merely to reduce turns, question count, or cognitive load.
- Professional language may expose the complete structured information contract at once and accept a batch answer. Business-guided language uses plain-language groups in impact order for as many turns as needed. When evidence is insufficient, show a neutral coverage outline and ask the user's preference. Every mode maintains the same complete question map; never omit a model-changing item merely to reduce turns, question count, or cognitive load.
- End every first response with the two-sentence fixed fact-gate footer defined by the requirement-clarification framework. Do not paraphrase, shorten, merge, or omit either sentence.
- End every first response with the two-sentence fixed fact-gate footer defined by the requirement-clarification framework. Do not paraphrase, shorten, merge, or omit either sentence.
@@ -82,25 +83,43 @@ Turn natural-language workforce-efficiency requests into reviewable and recovera
- Treat relative business-time expressions such as today, tomorrow, the next N days, or this quarter as `pending_confirmation` until the current request confirms the business timezone, business calendar, date-time anchor, and applicable day-boundary, cutoff, holiday, and overnight rules. A system clock or host timezone is environment evidence, not a business rule.
- Treat relative business-time expressions such as today, tomorrow, the next N days, or this quarter as `pending_confirmation` until the current request confirms the business timezone, business calendar, date-time anchor, and applicable day-boundary, cutoff, holiday, and overnight rules. A system clock or host timezone is environment evidence, not a business rule.
- Isolate every new request from historical contamination. Facts from another conversation, Plan, artifact, memory, or sample remain `pending_confirmation` until the user explicitly carries them into the current request; do not search for or reuse a semantically similar Plan as evidence for the first response.
- Isolate every new request from historical contamination. Facts from another conversation, Plan, artifact, memory, or sample remain `pending_confirmation` until the user explicitly carries them into the current request; do not search for or reuse a semantically similar Plan as evidence for the first response.
- Use one entry Agent for routing, consolidated questions, and final delivery. Assign one owner to each stage.
- Use one entry Agent for routing, consolidated questions, and final delivery. Assign one owner to each stage.
- Clarify-only requests still use this workflow. Generic AskUserQuestion may handle non-modeling setup choices or an explicitly explained capability fallback, but its replies never become DecisionWorkspace answers or human confirmations.
- Before using the nonexpert workflow, inspect the live DecisionWorkspace schemas for semantic_checks_version, request_clarification, interpret_clarification, sourceField and evidenceLinks, and confirm check_semantics is available. If the client lacks these contracts, remain in clarification and explain that a compatible client is required. Never emulate missing gates by editing AgentFS records or using the personal solver path.
- Preflight with a read-only DecisionWorkspace list call. If global TaskBoard/assistance capability is disabled or unavailable, explain the missing user interaction surface and remain in plain-text clarification; do not create questions the user cannot answer, and do not enable capabilities without their request.
- Clarification itself may create a team workspace with semantic_checks_version=1, propose business nodes, inspect an explicitly authorized data source, and record request_clarification before the fact gate passes. These are clarification operations, not permission to build a mathematical model, publish modeling artifacts, delegate solving or call a solver.
- Ask model-changing gaps through request_clarification with an answerable question, reason, canonical type, declared unit and justified constraints. Keep the same logical question while its statement/constraints remain unchanged. Do not invent bounds to reduce choices. Group questions by business impact and explain the remaining coverage in ordinary language.
- The platform already offers unknown and raw_text response channels. Do not add epistemic placeholders such as "not sure yet" or "unknown" to exact choices or node.value. These describe missing knowledge, not a business option. Explain the reason without repeating the UI's "why ask" label.
- When the current user message uniquely supplies a complete business value and every required unit/window, proposals must include a typed `node.value` and remain proposed until the dedicated human confirmation. An Agent-proposed value is not an Agent confirmation. Omitting value incorrectly downgrades a known fact to needs_input; a summary-only node is not a confirmable statement. Omit value only for genuinely unknown or still-ambiguous facts and ask a clarification instead.
- Human value answers are proposed, not confirmed. Use interpret_clarification only if raw_text uniquely determines the business value and every necessary unit/time scope. Preserve the original wording and link the normalized node to that answer; the user must explicitly confirm the new statement. If ambiguity remains, keep the original answer and needs_input, explain the remaining gap, and do not call interpret_clarification or fill a canonical placeholder. For example, "about eight hours, not sure per day or week" determines neither an exact limit nor a time window. Never use ordinary upsert to detach the original answer, manufacture a receipt, or default unknown.
- Let the platform continue the original source conversation after its effective gaps settle. Do not poll by starting new messages, change request identities to evade deduplication, revive canceled sources or retry unknown dispatch outcomes. A pending/interpretation signal is not solve authorization; reread the exact workspace and honor all gates.
- Bind real data with an authorized FileResourceRef, bytesHash, exact fields/types/units and confirmed timeScope. Add finite required/unique/range/enum/foreign_key/cutoff rules as applicable. Range/enum constants and FK fields must use consistent units. Every consumed data field must have modelSemantics.sourceField plus a sourced_from relationship before its business quantity is traced into the model. Unused columns need not be modeled. Unit/window conversions require an explicitly transformed, traceable and reconfirmed source; do not change only its unit label.
- Propose blocking validation nodes with confirmed positive, negative and boundary examples for exact model constraints. Include complete candidate assignments and the intended hard/soft behavior. Cover missing constraints, wrong time aggregation and accidental softening. Platform check_semantics independently evaluates the declared examples; a solver result or a second Agent explanation alone does not establish business correctness.
- Every new decision-grade `OptimizationSpec` must use the live TeamArtifact payload branch whose `schema_version` is exactly `2`; legacy v1 payloads are read-only compatibility artifacts and cannot be linked to approve a current decision. Before publishing, read the v2 schema and construct its complete algebraic/CP `model`, `problem_family`, exact `data_refs`, and `semantic_contract`. The contract must declare `solve_intent` and exactly one entry for every material objective, variable, constraint, and data reference actually present, with the confirmed dimension, canonical unit, and time scope where applicable—no missing, unknown, duplicate, display-label, JSON-Pointer, or implicit-unit substitutions. Re-read the published exact revision and verify this closed set before requesting `link_artifact`; a compile-success receipt for legacy or incomplete semantics is not decision-grade evidence.
- When peer review requests changes to a completed artifact stage, the original owner must reclaim the same canonical Task and Work Context, keep the old stage completed, and call `TeamArtifact(action="begin_rework")` with that completed source stage and the same stable artifact ID. Use only the platform-returned rework stage ID to publish the next revision of that same artifact, then complete the rework stage and resubmit the Task for a new review round. Never reopen a completed stage, create an ordinary replacement stage, switch to a new artifact ID, reuse a prior round's rework stage, or manufacture review/claim/Plan fences; if `begin_rework` is unavailable or rejected, stop and report the missing governance fact instead of publishing.
- Before confirmation and artifact linking, every actual `OptimizationSpec.data_refs` entry must have a confirmed business-source node whose `modelImpact` is `input` and whose dimension, canonical unit, and optional time scope exactly match the corresponding `semantic_contract` element. Data snapshots with `modelImpact=none` are display evidence, not material trace coverage. Positive/negative/boundary examples belong to deterministic semantic checks and must not be added as duplicate artifact material bindings.
- If a confirmed business change advances the model-input hash, refresh every consumed parent artifact under that exact binding before publishing its child; do not combine a fresh OptimizationSpec with stale SceneSpec/DataContract parents. After an interrupted/failed attempt, call `begin_rework` again with the same source stage and artifact. The platform may resume the existing stage only when the old Run is no longer running, the same review event/Task/Work Context/Plan remains authoritative, the new claim is strictly newer, and no new artifact revision was committed. Use the returned stage receipt; never invent a second stage to escape a stale or partial chain.
- Attach file evidence with evidenceLinks whose ref matches the node's evidenceRefs; use only authorized FileResourceRef identities, not guessed paths or arbitrary URLs. Explain report failures using their reason code, exact field/model element, frozen expectation, observed value, impact and next safe action. Direct the user to the graph's issue/source/model controls and change-impact preview; a colored branch, viewed evidence or preview never grants execution approval.
- After the fact-confirmation gate passes, follow this governed sequence without skipping or reordering its control points:
- After the fact-confirmation gate passes, follow this governed sequence without skipping or reordering its control points:
1. The entry Agent reads the needed `DecisionWorkspace(action="schema")` sections and each `TeamArtifact(action="schema")` contract before constructing writes.
1. The entry Agent reads the needed `DecisionWorkspace(action="schema")` sections and each `TeamArtifact(action="schema")` contract before constructing writes.
2. The entry Agent creates the team DecisionWorkspace or submits CAS-protected proposals in business language; proposals never manufacture human receipts.
2. The entry Agent creates the team DecisionWorkspace or submits CAS-protected proposals in business language; proposals never manufacture human receipts.
3. The user confirms or rejects blocking facts only through the platform's dedicated human controls. A rejection remains an auditable inactive tombstone. All Agents wait for the authoritative result.
3. The user confirms or rejects blocking facts only through the platform's dedicated human controls. A rejection remains an auditable inactive tombstone. All Agents wait for the authoritative result.
4. Only the top-level entry Agent calls `DecisionWorkspace(action="bind_workspace")` for the validated current workspace revision and model-input hash. Specialists may not bind, replace, or bypass it.
4. Only the top-level entry Agent calls `DecisionWorkspace(action="bind_workspace")` for the validated current workspace revision and model-input hash. Specialists may not bind, replace, or bypass it.
5. The assigned stage owner publishes`SceneSpec`, `DataContract`, optional `PredictionArtifact`, `OptimizationSpec`, `SolveResult`, `ValidationReport`, and `DeliveryBundle` in dependency order through `TeamArtifact`, retaining the returned exact artifact revision and DecisionWorkspace snapshot.
5. The assigned stage owners publish `SceneSpec`, `DataContract`, optional `PredictionArtifact`, and a decision-grade OptimizationSpec v2 with the complete `semantic_contract` above in dependency order through `TeamArtifact`, retaining exact artifact revisions and the DecisionWorkspace snapshot. When confirmed predictive inputs require training data, call OptimizationPredict inside this step, publish its real PredictionArtifact and confirm its model-changing outputs before creating OptimizationSpec. Without predictive inputs, omit this branch; do not predict after semantic checks. Initial model publication performs the platform's internal pure compilation checks; do not call guarded OptimizationCompile early to break the model-publication/semantic-check dependency.
6. The entry Agent calls `DecisionWorkspace(action="link_artifact")` with the exact artifact ID, exact revision, and semantic bindings; never resolve governed evidence through `latest`.
6. The entry Agent calls `DecisionWorkspace(action="link_artifact")` with the exact artifact ID, exact revision, and semantic bindings; never resolve governed evidence through `latest`.
7. Submit the linked revision for peer review. An independent reviewer checks business-to-model coverage, units, variable families, feasibility logic, provenance, gaps, and stale/rejected exclusions before execution.
7. Submit the linked revision for peer review. An independent reviewer checks business-to-model coverage, units, variable families, feasibility logic, provenance, gaps, and stale/rejected exclusions before execution.
After the user confirms the exact source/rules/examples, call DecisionWorkspace(action="check_semantics") for the current revision/hash and inspect its complete bounded report. Missing data, unsupported evaluators, budget limits or unconfirmed examples are blocked, not passes. Do not rebind the workspace merely because an audit-only revision changed; retain the model-input snapshot until business semantics change.
8. Invoke `OptimizationCompile` and then `OptimizationSolve`; both must pass the platform's DecisionWorkspace execution guard before side effects. `MindOptSolve` remains a compatible connector name and must never be called directly to bypass the guarded solve path.
8. Invoke `OptimizationCompile` and then `OptimizationSolve`; both must pass the platform's DecisionWorkspace execution guard before side effects. `MindOptSolve` remains a compatible connector name and must never be called directly to bypass the guarded solve path.
9. A validation owner independent from the solver owner calls `OptimizationValidate` and recomputes domains, hard constraints, objective, baseline delta, and IIS traceability from raw values.
9. A validation owner independent from the solver owner calls `OptimizationValidate` and recomputes domains, hard constraints, objective, baseline delta, and IIS traceability from raw values.
Only then publish and link the real SolveResult and ValidationReport. Never prepublish guessed results. DeliveryBundle must preserve the exact evidence chain and its human approval status.
10. The user approves only through the platform's dedicated human approval control after the exact linked chain passes review and independent validation. No Agent or specialist may create that approval.
10. The user approves only through the platform's dedicated human approval control after the exact linked chain passes review and independent validation. No Agent or specialist may create that approval.
-When training data exists, call`OptimizationPredict`; use ordered holdout, train-only imputation, tuning, metrics, baseline comparison, and explicit fallback rules.
-In step 5's confirmed prediction branch, use`OptimizationPredict` with ordered holdout, train-only imputation, tuning, metrics, baseline comparison, and explicit fallback rules.
- Compile general models with `OptimizationCompile`, solve once through guarded `OptimizationSolve`, and retain status, variables, objective, request/job IDs, HTTPS transport, selected-engine evidence, and IIS when infeasible.
- Compile general models with `OptimizationCompile`, solve once through guarded `OptimizationSolve`, and retain status, variables, objective, request/job IDs, HTTPS transport, selected-engine evidence, and IIS when infeasible.
- Require the validation owner to call `OptimizationValidate` and recompute variable domains, hard constraints, objective, baseline delta, and IIS traceability from raw values.
- Require the validation owner to call `OptimizationValidate` and recompute variable domains, hard constraints, objective, baseline delta, and IIS traceability from raw values.
- Treat every settled successful Tool call as authoritative. After an interruption, synthesize the persisted result without repeating the Tool.
- Treat every settled successful Tool call as authoritative. After an interruption, synthesize the persisted result without repeating the Tool.
## L2
## L2
- Until the fact-confirmation gate passes, clarify only: do not publish modeling artifacts, call a solver, delegate a solver-capable Agent, or fill hidden conditions with defaults, simulations, or industry convention.
- Until the fact-confirmation gate passes, use only the clarification operations above: do not publish modeling artifacts, call a solver, delegate a solver-capable Agent, or fill hidden conditions with defaults, simulations, or industry convention. Ledger labels such as confirmed_user are not substitutes for dedicated platform confirmation receipts.
- Interaction mode changes terminology, grouping, per-turn batch size, and example depth only. It never changes fact states, mandatory information, triggered conditions, the modeling-confirmation summary, or the stop gate. The user's explicit choice in the current request overrides historical preference; switching mode preserves confirmed facts and continues with every remaining item.
- Interaction mode changes terminology, grouping, per-turn batch size, and example depth only. It never changes fact states, mandatory information, triggered conditions, the modeling-confirmation summary, or the stop gate. The user's explicit choice in the current request overrides historical preference; switching mode preserves confirmed facts and continues with every remaining item.
- If the user cannot confirm an item, deliver the gap, model impact, required owner/data, and optional reduced scope. Never interpret an omitted answer as absent, false, zero, or unlimited.
- If the user cannot confirm an item, deliver the gap, model impact, required owner/data, and optional reduced scope. Never interpret an omitted answer as absent, false, zero, or unlimited.
- Use the fast path only when the user supplied a complete `OptimizationSpec`. It may omit unnecessary scene, prediction, or data-authoring work, but it must still create/propose the governed DecisionWorkspace representation, obtain every required dedicated human confirmation, bind the current revision, publish and link exact revisions, pass peer review and execution guards, delegate one solve and one independent validation, and wait for dedicated human approval before an entry-owned `DeliveryBundle` is decision-grade.
- Use the fast path only when the user supplied a complete `OptimizationSpec`. It may omit unnecessary scene, prediction, or data-authoring work, but it must still create/propose the governed DecisionWorkspace representation, obtain every required dedicated human confirmation, bind the current revision, publish and link exact revisions, pass peer review and execution guards, delegate one solve and one independent validation, and wait for dedicated human approval before an entry-owned `DeliveryBundle` is decision-grade.
- 在真人确认与工件关联前，`OptimizationSpec.data_refs` 中每个实际数据引用都必须有一个已确认业务来源节点，其 `modelImpact` 明确为 `input`，dimension、canonical unit 及可选 time scope 与 `semantic_contract` 对应元素完全一致。`modelImpact=none` 的数据快照只是展示证据，不能覆盖 material trace；正例、反例和临界例属于确定性语义检查，不得重复加入artifact material bindings。
"summary":"Co-manage workforce models, guarded solving, and independent validation; MindOpt is externally licensed",
"summary":"Co-manage workforce models, guarded solving, and independent validation; MindOpt is externally licensed",
"description":"When users naturally request service coverage, hierarchical resource allocation, performance targets, centralized task scheduling, staffing, shift planning, or general LP/MILP, use DecisionWorkspace humanconfirmation, exact-version artifacts, entry Agent binding, peer review, execution guards, independent validation, and human approval to create a recoverable governance chain. When MindOpt is selected, actual solving still requires the user to obtain an applicable license, deploy MindOpt, and configure its connector; the client and this Skill do not include the solver, license, hosted compute, or related fees."
"description":"Load this Skill for the first clarification of staffing, shifts, work-hour limits, service coverage, resource allocation, targets, task scheduling or LP/MILP, including clarify only / do not solve yet requests. Use DecisionWorkspace question and human-confirmation cards for model-changing facts; generic AskUserQuestion is not a substitute. Clarification needs no connected solver. Actual MindOpt solving requires a separately licensed/deployed solver and configured connector; software, licenses, hosting and fees are not included."
@@ -17,6 +17,8 @@ Maintain `fact_state`, `value`, `source`, and `model_impact` for every modeling
Mandatory rules:
Mandatory rules:
The ledger states describe evidence for discussion, not authorization receipts. In the governed DecisionWorkspace, model-changing statements still require dedicated human confirmation. Before that gate, schema discovery, business-node proposals, authorized data inspection and request_clarification are permitted clarification actions; mathematical modeling/artifact publication/solver delegation remain prohibited. Use interpret_clarification only when raw_text uniquely determines the business value and required unit/window, then obtain explicit confirmation. Ambiguous raw_text stays needs_input with its original receipt; explain the gap without inventing a normalized value. Unknown uses the platform's unknown channel, not an epistemic placeholder in exact choices or node.value. Conversation text, this ledger and Agent-owned files cannot manufacture platform human facts.
1. Never interpret a missing answer as absent, false, zero, unlimited, or industry standard.
1. Never interpret a missing answer as absent, false, zero, unlimited, or industry standard.
2. Never replace confirmation with a reasonable default or a simulated possible situation.
2. Never replace confirmation with a reasonable default or a simulated possible situation.
3. If a factor has two or more model-changing values, mark it `pending_confirmation` and ask.
3. If a factor has two or more model-changing values, mark it `pending_confirmation` and ask.
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
GitHub