medication-information-assistant — Inspect Lite 检测报告
报告头(Provenance)
- Host model:自声明宿主模型(本会话执行模型,未提供具体型号标识)
- Skill package:
sanityops-inspect-lite v1.1.0 - Specification baseline:
SanityOps Framework v1.1 (2026-10 release) - 非官方结果声明:本结果由宿主模型执行的技能规则包产生,非官方 SanityOps 引擎结论;完整检测请使用
sanityops-cli(https://github.com/sanityops-org/sanityops-framework/blob/main/framework/try-the-tools.md)。 - 检测日期:2026-11-04
- 工件版本标签:
undeclared(来源文件samples/cases/medication-information-assistant.md,无显式版本标签) - Fidelity(保真度):
full(全部为完整 verbatim 工件,非摘要;按 full 运行全部规则)
输入说明(Object Segmentation 结果)
来源文件是「容器」文档,内嵌多个 verbatim 逻辑工件。按 SKILL.md §2 的对象切分:
- 工件段(受检对象):
- System Prompt —
prompts/system.txt(第 49–239 行 fenced block) - Skill —
skills/medication_info/SKILL.md(第 247–433 行) - Tool Schema ×5 —
medication_search、medication_get_usage、medication_check_interactions、medication_get_storage、medication_get_precautions(第 445–598 行,各parametersdict + docstring)
- System Prompt —
- 容器脚手架(不检测):front matter、目录(Table of Contents)、Report Metadata 表、PASS Criteria、「Rule-version note」、各 section 标题、「Report Generated」页脚。
- 层级 L:Prompt = AC-L2(元数据声明 "Inspection Level: L2",多轮对话 + 工具调用);Skill 与 Tool = OR-L1(全部只读操作,无写/删/外发)。
- Tool Schema 边界:各 Tool 的
execute()正文被省略("for brevity")——非 schema 部分,不构成缺陷;parametersdict + docstring 即完整 Tool Schema。 - QC 范围:仅检内嵌于 Prompt/Skill/Tool 文本字段内的代码片段。本用例仅 Skill 的「## Usage Example」python 代码块为 QC 对象,按「示意片段」规则处理(示意性示例豁免 QC-0.1/QC-0.3 完整性要求,省略/缺 import 不构成 FAIL)。其余工件无内嵌代码。
执行摘要(Executive Summary)
- 已执行阶段:Stage A / B / C / D / E / F / G 全部执行(七阶段非阻塞流程,P0 聚合上报,未中途停止)。
- P0/P1/P2 计数:P0 = 1 · P1 = 1 · P2 = 0(另有 Skill 基线门 2 项未通过、参考级 advisory 若干,均不计入 P 计数)。
- Gate-0 结果:G0-4 = complete(准入通过)→ Stage D 执行;G0-1 携带 Stage A P0(QD-P-2.5.6);G0-2 无 Stage B P0;G0-3 = locatable。
- Explicit·P0/P1/P2 摘要:0 / 0 / 0 —— Stage E 扫描 34 个 EX 子类,未发现任何显式风险表达。
- Top findings(≤5):
QD-P-2.5.6— Instruction-Data Trust Boundary(指令-数据信任边界)未声明,Prompt 缺少防注入声明(P0)。QD-P-2.4.4— 多轮上下文/记忆边界未在 Prompt 中声明(P1)。QD-S-0.5— Skill 中medication_check_interactions的 Returns 清单存在重复/冲突字段(基线门未通过)。QD-S-0.8— Skill front matter 缺少 version/owner/changelog(基线门未通过)。QD-ST-3.5(参考级)— Skill 声明的输出结构与 Tool 实际返回不一致(缺medications_checked/foods_included等)。
- 未运行项:Risk Implicit(动态可利用性)、运行时 Quality 子集(RAG-Agent/Tool-Agent 运行时评测)。lite 只运行 Stage A 的内嵌代码 QC 与 Stage E 的 Risk Explicit。
- 升级提示:完整 Inspect 规则集 + Risk(Explicit/Implicit)+ Quality 子集、打分、版本级证据链,请使用
sanityops-cli。
本报告不给出任何安全或质量结论(Inspect PASS ≠ Security PASS ≠ Quality PASS)。
术语速查(Terminology Quick Reference)
| ID 家族 | 含义 | 白话解释 |
|---|---|---|
QD-P* / QD-S* / QD-T* | Inspect 缺陷 ID | Prompt / Skill / Tool Schema 内的单工件缺陷 |
QD-PS/PT/ST/SS/TT/CYC | 跨工件缺陷 ID | 两个工件之间的一致性问题 |
QD-PM | 权限比例性缺陷 ID | 声明权限与职责是否成比例的问题 |
QC-* | 内嵌代码缺陷 ID | 工件内嵌代码片段的缺陷 |
EX-* | Risk Explicit ID | 静态文本中显式表达的风险 |
AS-* / FM-* | 影响映射 ID | 候选攻击面 / 候选失效模式(仅建议性关联) |
AC-L1/L2/L3 | Agent 复杂度层级 | 决定 Prompt 域 severity 的复杂度档位 |
OR-L1/L2/L3 | 操作风险层级 | 决定 Skill/Tool 域 severity 的风险档位(L1 只读 / L2 可逆写 / L3 不可逆) |
P0 / P1 / P2 | 严重度 | 阻断 / 警告 / 建议 |
full / summary | 保真度 | 原始文件(全规则)vs 摘要(仅矛盾类规则) |
applicability.fact | 结构事实 | 决定某项检查「考不考」的可观测事实(如 has_tool) |
| 参考级 | reference level | 统一清单携带、但原 tier 规范在该 L 下不分配 P 的项——照查,但作为 advisory 展示并标注原始 severity |
Stage A — 单工件检测
逐条 verdict 格式:
ID | PASS/FAIL/N/A | 证据位置。FAIL 引原文短句;N/A 带applicability.fact结构依据。L 只决定 severity,不决定「考不考」。
A1. QD-P — System Prompt(AC-L2)
| ID | 判定 | AC-L2 severity | 证据 / 依据 |
|---|---|---|---|
| QD-P-1.1.1 Direct Contradiction | PASS | P0 | 全文无直接矛盾;优先级、规则、流程相互一致 |
| QD-P-1.1.2 Priority Conflict | PASS | P1 | 显式仲裁:# Priority Order "The first matching rule wins; later rules never override earlier ones" |
| QD-P-1.1.3 Example-Rule Contradiction | PASS | P0 | Safety Examples 与 Rule 1/2/3 一致(❌/✓ 示例与固定回复相符) |
| QD-P-1.2.1 Role-Responsibility Mismatch | PASS | P0 | "You are not a doctor or pharmacist" 与信息提供职责匹配 |
| QD-P-1.2.2 Role-Output Style Mismatch | PASS | P1 | 输出风格(plain language, no jargon)与教育型角色一致 |
| QD-P-1.2.3 Constraint-Responsibility Mismatch | PASS | P0 | 禁处方/禁剂量约束落在正确的职责范围 |
| QD-P-1.2.4 Identity & Anthropomorphism | PASS | P1 | 明确 AI 身份、无冒充真人("You are the Medication Information Assistant") |
| QD-P-1.3.1 Quantifier Misuse | PASS | P1 | "exactly once"、"at most one"、"exactly five" 量词精确 |
| QD-P-1.3.2 Vague Qualifiers | PASS | P1 | 无模糊决策边界词 |
| QD-P-1.3.3 Undefined Boundaries | PASS | P0 | Task/Emergency/Boundary Statement 边界定义完整 |
| QD-P-1.4.3 Invalid Skill Definitions | PASS | P1 | 唯一 Skill medication_info 定义清晰、可触发 |
| QD-P-1.5.1 Responsibilities Not Support Objective | PASS | P0 | 职责集合支撑目标 |
| QD-P-1.5.2 Resources Not Support Workflow | PASS | P0 | 工作流引用的 5 工具 + 2 数据文件均已定义 |
| QD-P-1.6.1 Critical Decision Word Vagueness | PASS | P1 | "Relevance means named in the user's current question" 已定义 |
| QD-P-1.6.2 Rules Depending on Model Internal State | PASS | P1 | 决策依据均为可观测事实(搜索命中数、错误返回) |
| QD-P-1.6.3 Soft Rule Interweaving Without Arbitration | PASS | P1 | 优先级仲裁显式存在 |
| QD-P-1.6.4 Reasoning Method Constraints | PASS | P1 | 无不可验证推理方法作保证 |
| QD-P-1.6.5 Self-Referential/Recursive Constraints | PASS | P1 | 无自指/递归约束 |
| QD-P-1.6.6 Implicit Assumption Dependency | PASS | P1 | 依赖(内置参考库)已显式声明 |
| QD-P-1.6.7 Non-Monotonic Reasoning Defects | PASS | P1 | "Identical inputs plus identical tool results produce an identical output structure" 确定性声明 |
| QD-P-1.6.8 Rule Followability | PASS | P1 | "respond exactly"/"verbatim" 规则可执行 |
| QD-P-2.1.2 Reference Clarity | PASS | P1 | Authorized Tools 表 + Invocation Specification 引用清晰可定位 |
| QD-P-2.2.1 Undeclared Input Source | PASS | P0 | "Most inputs come from the user's message" 输入来源已声明 |
| QD-P-2.2.2 Unconstrained Input Format | PASS | P0 | "plain text in UTF-8, no longer than 2,000 characters" 格式/尺寸已声明 |
| QD-P-2.2.3 Required/Optional Not Distinguished | PASS | P0 | "Required and Optional Inputs" 小节区分必填/可选 |
| QD-P-2.2.4 Undefined Missing-Input Handling | PASS | P0 | "Missing Input Handling" 小节已定义 |
| QD-P-2.3.1 Undefined Output Format | N/A | P0 | has_machine_consumed_output 不成立:输出为面向患者的 Markdown 对话,非机器消费 |
| QD-P-2.3.2 Incomplete Output Schema | N/A | P0 | has_structured_output 不成立:输出为 Markdown 文本段落,非结构化/JSON schema |
| QD-P-2.3.4 Missing Security/Privacy Filtering | PASS | P0(条件) | has_pii 成立(健康域);已声明过滤:"Never include... Personal health details the user shares... do not repeat them back" |
| QD-P-2.3.5 Factuality & Hallucination Control | PASS | P0 | "Never fabricate medication names or medication_ids. Only use IDs returned by medication_search." |
| QD-P-2.3.6 Output Length & Truncation | PASS | P2 | "Keep the Answer to 150 words... interaction-check answers may be 400 words" |
| QD-P-2.3.7 Output Localization & Conventions | N/A | P1(条件) | is_multi_region 不成立:单区域场景 |
| QD-P-2.4.1 Incomplete Workflow Steps | PASS | P1 | Conversation Flow 三步完整有序 |
| QD-P-2.4.2 Undeclared Resources/Tools | PASS | P0 | 5 工具 + 2 数据文件全部声明 |
| QD-P-2.4.3 Tool Invocation Parameter/Order Error | PASS | P1 | 参数与 schema 一致(name 2-100、limit 1-20、med_+3 digits 等) |
| QD-P-2.4.4 Multi-turn Context & Memory Boundary | FAIL | P1 | 证据:Prompt 正文未声明跨轮/跨会话上下文与记忆边界("Relevance means named in the user's current question" 仅约束当前问题相关性,未声明是否/如何使用会话历史或跨轮记忆)。该声明存在于 Skill(另一工件段)——"Do not access conversation history for data"——但 Stage A 单工件边界下不得用兄弟段替 Prompt 段满足 |
| QD-P-2.5.1 Missing Termination Constraints | PASS | P1 | "After 12 total calls, summarize instead of calling more tools" 终止约束已声明 |
| QD-P-2.5.2 Missing Security Boundary Constraints | PASS | P0 | Emergency Safety Boundary + Three Safety Rules + Security/Privacy Filtering + Boundary Statement |
| QD-P-2.5.3 Missing Permission Control Boundaries | PASS | P1 | "exactly five read-only tools... none creates, updates, or deletes anything" |
| QD-P-2.5.4 Undefined Constraint Priority | PASS | P1 | Priority Order 显式定义仲裁 |
| QD-P-2.5.6 Instruction-Data Trust Boundary | FAIL | P0 | 证据:has_external_data 成立(接收用户输入 + 工具返回),但 Prompt 未声明「外部输入/工具返回仅作数据、不得作指令」的信任边界(防注入)。现有语句("Only use IDs returned by medication_search")是数据完整性约束,非指令-数据信任边界 |
| QD-P-2.5.7 High-Risk Operation Human Confirmation | N/A | P0(条件) | has_side_effect_op 不成立:全部工具只读,无资金/删除/发送/生产变更等高危操作 |
| QD-P-2.5.8 System Prompt Confidentiality & Leakage | N/A | P1 | has_confidential_content 不成立:Prompt 本身不含机密内容 |
| QD-P-2.5.9 Industry Compliance & Disclaimer | PASS | P0(条件) | is_regulated_domain 成立(医疗域);已声明资质边界("not a doctor or pharmacist")+ 强制免责声明(Rule 3) |
| QD-P-2.6.1 Runtime Exceptions Not Covered | PASS | P1 | "On error, tell the user the lookup failed... Do not retry the identical call." |
| QD-P-2.6.2 Missing Degradation & Recovery | N/A | ∅(参考,L3=P1) | is_complex_agent 不成立(声明层级 AC-L2,非复杂 Agent)。注:Stage G 层级自审见下 |
| QD-P-2.6.3 Unreasonable Retry Mechanism | PASS | ∅(参考,L3=P1) | "Do not retry the identical call" + "After 12 total calls, summarize" 重试机制有界合理 |
| QD-P-2.6.4 Out-of-scope Rejection & Escalation | PASS | P1 | 拒绝模板 + 转介(pharmacist)+ 升级(Emergency)已定义 |
建议项(ungraded,不评分不阻断):QD-P-1.4.1 / 2.1.3 / 2.1.5 / 2.7.1 / 2.7.2 / 2.7.3 —— 均为
is_prompt优化类,不计入本报告 verdict。 Tombstone:QD-P-2.1.1(deleted)、2.1.4(merged→1.4.1)、2.5.5(deprecated)、2.3.3(编号空位)——不判定。
A2. QD-S — Skill(OR-L1,先跑 QD-S-0 基线 9 项)
QD-S-0 基线门(9 项,pass/fail,不计分):
| ID | 判定 | 证据 / 依据 |
|---|---|---|
| QD-S-0.1 Required Field Presence | PASS | name/description 均存在 |
| QD-S-0.2 Field Content Non-Empty | PASS | 各字段非空 |
| QD-S-0.3 Syntax Parsability | PASS | YAML front matter 可解析 |
| QD-S-0.4 Field Type Validity | PASS | 字段类型合法(list/string/bool/int) |
| QD-S-0.5 No Conflicting Declarations | FAIL | 证据:medication_check_interactions 的 Returns 块重复声明同一字段且相互冲突——先列 - interactions_found: Boolean... 与 - interactions: List of identified interactions (empty list when none found), each with only: severity/type/description/recommendation,随后再次列出 - interactions_found: Boolean... 与 - interactions: List of found interactions (see below),并额外出现 summary。同名字段两次定义、第二次语义与第一次不一致 |
| QD-S-0.6 Minimum Structural Completeness | PASS | 可定位最小结构(front matter + 分节 body) |
| QD-S-0.7 Dependency & Resource Existence | PASS | 引用的 5 工具 schema 存在、2 数据文件声明;无脚本引用 |
| QD-S-0.8 Metadata Completeness | FAIL | 证据:front matter 缺少 version / owner / changelog 字段(第 248–283 行仅含 name/description/invocation/allowed_operations/allowed_resources/side_effects/non_goals/data_sensitivity/partial_result_allowed/max_retries/max_tool_calls/timeout_seconds/tools) |
| QD-S-0.9 description Length Budget | PASS | description 短,未超预算(SHOULD 级) |
QD-S-0.3 通过 → 不阻断 QD-S-1~5;0.5/0.8 未通过 → 记录为「baseline not passed」,Skill 结论标注「assessment incomplete (NEEDS_REVIEW)」。
QD-S 语义项(24 项,OR-L1):
| ID | 判定 | OR-L1 severity | 证据 / 依据 |
|---|---|---|---|
| QD-S-1.1 Input Boundary Missing | PASS | P1 | 各操作参数声明了输入边界(name 2-100、limit 1-20、medication_ids 1-10) |
| QD-S-1.2 Output Size Missing | PASS | P1 | "Output Limits — Limit each response to at most 150 words..." |
| QD-S-1.3 Execution Resource Boundary Missing | PASS | ∅(参考,L2/L3=P0) | max_retries:3、max_tool_calls:12、timeout_seconds:60 |
| QD-S-1.4 Permission Boundary Missing | PASS | ∅(参考,L2=P1/L3=P0) | allowed_operations + "This skill may invoke only the five tools... Invoking any other skill or tool is prohibited." |
| QD-S-1.5 Data Source Boundary Missing | PASS | ∅(参考) | "Data Source" 节 + "Never access production databases... use only the built-in local reference data" |
| QD-S-2.1 Unbounded Quantity Declaration | PASS | P1 | 无 all/every/entire/complete 无界量词 |
| QD-S-2.2 Unterminated Execution Declaration | PASS | ∅(参考) | "max_retries: 3"、"stop immediately";无 keep trying/until success |
| QD-S-2.3 Open-Ended Enumeration in Structure | PASS | ∅(参考) | Returns 显式 "closed enumeration - do not add extra fields";无 etc./such as 开放枚举 |
| QD-S-2.4 Broad Capability Verbs | PASS | ∅(参考) | 动词具体(search/retrieve/check),无 manage/handle 宽泛动词 |
| QD-S-3.1 Vague Trigger Conditions | PASS | P1 | "Trigger this skill whenever the user names a product..." + When to Use / Do NOT use for 清单 |
| QD-S-3.2 Missing Deactivation Conditions | PASS | ∅(参考;L1 可选) | "Do NOT use for:" 清单已声明 |
| QD-S-3.3 Missing Ambiguous Input Handling | PASS | P1 | "when no name is given or the brand-vs-generic form is unclear, ask the user instead of guessing" |
| QD-S-3.4 Undefined Task Scope | PASS | P2 | non_goals + "Do NOT use for" 定义任务范围 |
| QD-S-3.5 Undeclared Default Behavior | PASS | ∅(参考) | "Partial results are acceptable... mark missing sections as 'information not available'" |
| QD-S-3.6 Metadata Expression Defects | PASS | P1 | name/description 与 body 一致(如实描述只读取药信息;返回清单重复问题归 QD-S-0.5) |
| QD-S-3.7 Output Structure Contract Missing | PASS | ∅(参考) | 各操作 Returns 块已声明输出结构(medication_check_interactions 的重复/冲突归 QD-S-0.5) |
| QD-S-4.1 Undefined Failure Behavior | PASS | P1 | "If a tool call fails or returns an error, stop immediately and report the error" |
| QD-S-4.2 Undeclared Partial Completion Legitimacy | PASS | ∅(参考) | partial_result_allowed:true + "Partial results are acceptable" |
| QD-S-4.3 Missing Post-Failure Prohibited Behaviors | PASS | ∅(参考) | "do not expand the task scope, fabricate data... guess medication_ids, or source information outside..." |
| QD-S-4.4 Error Recovery Path Undefined | PASS | ∅(参考) | "stop immediately and report the error"(safe-terminate-and-report 合法处置) |
| QD-S-5.1 Read/Write/Delete Not Distinguished | N/A | ∅ | has_write_semantics 不成立:side_effects:none,全部只读 |
| QD-S-5.2 Missing High-Risk Confirmation | N/A | ∅ | has_irreversible_op 不成立:无可逆/高危操作 |
| QD-S-5.3 Undeclared Environment Boundaries | PASS | ∅(参考) | "Never access production databases, production APIs, live systems... use only the built-in local reference data" |
| QD-S-5.4 Undeclared Chained Invocation Permissions | N/A | ∅ | has_skill_chaining 不成立:明确禁止调用其他 skill("Invoking any other skill or tool is prohibited") |
A3. QD-T — Tool Schema(5 个,OR-L1)
5 个 schema 均为「parameters dict + docstring」。四项(get_usage/get_storage/get_precautions/check_interactions)结构近似;逐项列 verdict,相同处合并说明。
| ID | medication_search | medication_get_usage | medication_get_storage | medication_get_precautions | medication_check_interactions | 依据(共性) |
|---|---|---|---|---|---|---|
| QD-T-1.1 Base Field Missing (P0) | PASS | PASS | PASS | PASS | PASS | name+description(docstring)+inputSchema(parameters) 齐全 |
| QD-T-1.2 Required Refs Undefined (P0) | PASS | PASS | PASS | PASS | PASS | required 字段均在 properties 定义 |
| QD-T-1.3 Enum conflict (P1·P2) | N/A | N/A | N/A | N/A | N/A | has_enum 不成立:无 enum 字段 |
| QD-T-1.4 additionalProperties Policy (P2) | PASS | PASS | PASS | PASS | PASS | 封闭契约字段,additionalProperties: False 恰当 |
| QD-T-1.5 name Validity & Uniqueness (P0) | PASS | PASS | PASS | PASS | PASS | 名称合法且唯一 |
| QD-T-1.6 Constraint Self-Consistency (P1) | PASS | PASS | PASS | PASS | PASS | min≤max、default 在范围内(limit 1–20 default 5) |
| QD-T-1.7 Type Missing/Over-Broad (P2) | PASS | PASS | PASS | PASS | PASS | 类型明确,无 type:any/嵌套>4/参数>30 |
| QD-T-2.1 String Without maxLength (P2) | PASS(name maxLength 100) | PASS(id maxLength 7) | PASS | PASS | PASS(items maxLength 7) | 字符串参数均有 maxLength |
| QD-T-2.2 Format Contract Missing (P2) | PASS | PASS | PASS | PASS | PASS | id/ids 有 pattern ^med_\d{3}$;name 无需特定格式 |
| QD-T-2.3 Number Without Boundary (P2) | PASS(limit min1 max20) | N/A | N/A | N/A | N/A | 仅 search 有 number;has_number_param 其余不成立 |
| QD-T-2.4 Array Without maxItems (P2) | N/A | N/A | N/A | N/A | PASS(maxItems 10) | 仅 check_interactions 有 array |
| QD-T-2.5 Unit/Precision/Timezone (P2) | N/A | N/A | N/A | N/A | N/A | has_amount_or_time_param 不成立:无金额/时间参数 |
| QD-T-2.6 Query Without Pagination (P1) | PASS(limit 参数) | N/A | N/A | N/A | N/A | is_query_tool 仅 search 成立,且有 limit cap |
| QD-T-2.7 Param Dependency (P1) | N/A | N/A | N/A | N/A | N/A | has_param_dependency 不成立:无依赖/互斥 |
| QD-T-3.1 State Change Undeclared (∅) | N/A | N/A | N/A | N/A | N/A | has_write_semantics 不成立:只读 |
| QD-T-3.2 Untrusted Input Executable (P1) | N/A | N/A | N/A | N/A | N/A | has_executable_param 不成立:无 SQL/shell/code 参数 |
| QD-T-3.3 Batch Write Without Cap (∅) | N/A | N/A | N/A | N/A | N/A | has_batch_write 不成立:无批量写 |
| QD-T-3.4 Authz Context Exposure (P0) | N/A | N/A | N/A | N/A | N/A | has_authz_param 不成立:无 tenantId/admin/token 参数 |
| QD-T-3.6 Dangerous Default (∅) | N/A | N/A | N/A | N/A | N/A | has_dangerous_default 不成立:默认值均安全 |
| QD-T-3.7 Op Whitelist (P0) | N/A | N/A | N/A | N/A | N/A | is_openended_tool 不成立:非 sql/exec/http/eval |
| QD-T-4.1 Insufficient Description Quality (P2) | PASS | PASS | PASS | PASS | PASS | docstring 覆盖动作/对象/返回摘要等 ≥2 要素 |
| QD-T-4.4 Param description Missing (P2) | PASS | PASS | PASS | PASS | PASS | 各 property 均有 description |
| QD-T-4.5 description Inconsistent w/ Schema (P1) | PASS | PASS | PASS | PASS | PASS | 描述与约束一致("1-10 IDs" vs minItems1/maxItems10 等) |
| QD-T-4.6 Sensitive Input Undeclared (∅) | N/A | N/A | N/A | N/A | N/A | has_sensitive_param 不成立:无凭证/密钥类参数 |
| QD-T-4.7 Output Contract Missing (outputSchema) | FAIL(参考级) | FAIL(参考级) | FAIL(参考级) | FAIL(参考级) | FAIL(参考级) | OR-L1 下 ∅(参考级);原始级 ∅/P1/P1→P0(cross-flow)。证据:仅 docstring 描述返回字段,无正式 outputSchema 字段。参考级 advisory,不计入 P |
| QD-T-5.1/5.2/5.3 Change-check | N/A | N/A | N/A | N/A | N/A | has_version_change 不成立:无 old/new schema 对比 |
QD-T-5.x 为 change-check(不计入静态分母)。Tombstone:QD-T-4.2(merged→3.1)、3.5(transferred→QD-TT-5.1)、4.3(transferred→QD-TT-5.2)——不判定。
A4. QC — 内嵌代码片段(Skill「## Usage Example」python 块)
该片段为「示意片段」(## Usage Example 标记,非正式可执行代码)。QC-0 基线先跑。
QC-0 基线门(5 项):
| ID | 判定 | 依据 |
|---|---|---|
| QC-0.1 Syntax Parsability (P0) | PASS | Python 语法可解析(示意片段豁免完整性) |
| QC-0.2 Language Declaration (P1) | PASS | fence 声明 ```python |
| QC-0.3 Structural Completeness (P1) | PASS | 示意片段豁免(缺 import 不构成 FAIL) |
| QC-0.4 Fence Integrity (P1) | PASS | 围栏开/闭匹配 |
| QC-0.5 Language-Content Mismatch (P2) | PASS | python fence 内容为 python |
QC-1~4(12 项):
| ID | 判定 | 依据 |
|---|---|---|
| QC-1.1 Undefined Name Reference (P1) | PASS | run 为框架工具调用原语,示意片段「缺 import 不构成 FAIL」豁免 |
| QC-1.2 Invalid Import (P2) | N/A | has_import 不成立:片段无 import |
| QC-1.3 Deprecated API Usage (P2) | Not Executed | 依赖上下文(API 版本库)缺失,lite 默认 Not Executed,绝不记 FAIL |
| QC-1.4 Type Mismatch (P2) | PASS | 字典/列表/布尔字面量与调用形参类型相容 |
| QC-2.1 Unclosed File Handle (P1) | N/A | has_file_handle 不成立 |
| QC-2.2 Unreleased Network Connection (P2) | N/A | has_network_conn 不成立 |
| QC-2.3 Missing Exception Handling (P2) | N/A | has_high_risk_op 不成立:无 file/network/DB 直接操作 |
| QC-3.1 Unused Import (P2) | N/A | has_import 不成立 |
| QC-3.2 Missing Import (P1) | PASS | run 未 import,但示意片段「缺 import 不构成 FAIL」豁免 |
| QC-3.3 Hardcoded Demo Value (P2) | PASS | "ibuprofen"/"med_001"/"med_005" 为示例药物数据,非弱凭证/机器特定值 |
| QC-4.1 Snippet-Artifact Declaration Mismatch (P1) | PASS | 调用与 allowed_operations 及 schema 参数一致(name/medication_id/medication_ids/include_foods) |
| QC-4.2 Snippet-Body Version/API Drift (P2) | PASS | 无版本/API 引用漂移 |
Stage B — 跨工件检测(29 项,AC-L2)
has_*为关系结构事实。单 Agent、单 Skill、5 Tool。
| ID | 判定 | AC-L2 severity | 证据 / 依据 |
|---|---|---|---|
| QD-PS-1.1 Skill Authorization Consistency | PASS | P0 | Prompt 声明 skill medication_info,已定义且授权一致 |
| QD-PS-1.2 Trigger Condition Consistency | PASS | P1 | Prompt「names a medication」触发 ↔ Skill「names a product they want medication information about」一致 |
| QD-PS-1.3 Permission Boundary Consistency | PASS | P0 | Prompt「five read-only tools」↔ Skill allowed_operations(5 工具)一致 |
| QD-PS-1.4 Resource Access Consistency | PASS | P1 | Prompt 资源 ↔ Skill allowed_resources(data/medications.json、data/interactions.json)一致 |
| QD-PS-1.5 Failure Handling Consistency | PASS | ∅(参考,L3=P1) | Prompt「report + do not retry」↔ Skill「stop and report」一致 |
| QD-PS-1.6 Output Constraint Consistency | PASS | ∅(参考,L3=P2) | Prompt 150/400 字 ↔ Skill「150 words」一致 |
| QD-PS-1.7 Orchestration Order & Precondition | PASS | P1 | Prompt「resolve name with medication_search first」↔ Skill「medication_id obtained from search」一致 |
| QD-PT-2.1 Tool Existence | PASS | P0 | Prompt 引用 5 工具,5 个 schema 均存在 |
| QD-PT-2.2 Invocation Specification Consistency | PASS | P0 | Prompt 工具表参数 ↔ schema(name/limit/medication_id/medication_ids/include_foods)一致 |
| QD-PT-2.3 Parameter Constraint Consistency | PASS | P1 | name 2-100、limit 1-20、ids 1-10、med_+3 digits 与 schema pattern 一致 |
| QD-PT-2.4 Permission Granularity Consistency | PASS | P0 | Prompt「read-only」↔ Tool 无写语义一致 |
| QD-PT-2.5 Key Error Handling Coverage | PASS | ∅(参考,L3=P1) | Prompt 覆盖工具错误(report + verify name) |
| QD-PT-2.6 Business-Level Frequency Expectation | PASS | ∅(参考,L3=P2) | Prompt 声明调用预算(2/tool/turn、12/conversation) |
| QD-PT-2.7 Parameter Semantics & Unit Consistency | PASS | P1 | med_+3 digits 语义与 schema pattern ^med_\d{3}$ 一致;无金额/单位参数 |
| QD-ST-3.1 Tool Existence | PASS | P0 | Skill allowed_operations/tools 引用 5 工具,均存在 |
| QD-ST-3.2 Input Constraint Consistency | PASS | P0 | Skill 参数约束(name 2-100、ids 1-10)↔ schema 一致 |
| QD-ST-3.3 Required Parameter Preparation | PASS | P0 | 必填参数均有来源声明(medication_id "from search results" 等) |
| QD-ST-3.4 Business-Level Frequency Expectation | PASS | ∅(参考,L3=P2) | Skill 声明 max_tool_calls/max_retries/2-per-turn |
| QD-ST-3.5 Output Structure Match | FAIL(参考级) | ∅(参考,L3=P1) | 证据:Skill 声明输出与 Tool 实际返回不一致——(a) medication_check_interactions:Skill Returns 缺 Tool docstring 返回的 medications_checked、foods_included,且重复 interactions_found/interactions;(b) medication_search:Skill 缺 Tool 返回的 total_found、search_term;(c) medication_get_usage:Skill 列 form,Tool docstring 未返回 form;(d) medication_get_storage:Skill 列 special_instructions,Tool docstring 未列。参考级 advisory,不计入 P |
| QD-ST-3.6 Permission Scope Match | PASS | P0 | Skill allowed_operations ↔ Tool 副作用(均只读)一致 |
| QD-ST-3.7 Key Error Handling Coverage | PASS | ∅(参考,L3=P1) | Skill Failure Handling 覆盖工具错误 |
| QD-CYC-1.1 Cross-Artifact Circular Invocation | PASS | P1 | Prompt→Skill→Tool 单向,无闭环 |
| QD-CYC-1.2 Multi-Agent Circular Invocation | N/A | ∅ | has_multi_agent 不成立:单 Agent |
| QD-CYC-1.3 Inter-Skill Handoff Loop | N/A | ∅ | has_skill_handoff 不成立:仅 1 个 Skill |
| QD-SS-4.1 Trigger Condition Conflict | N/A | ∅ | has_multiple_skills 不成立:仅 1 个 Skill |
| QD-SS-4.2 Name Uniqueness | N/A | ∅ | has_multiple_skills 不成立 |
| QD-TT-5.1 Cross-Tool Data Flow | PASS | P1 | medication_id 从 search 流向 get_*,无「敏感输出→危险消费」链(全只读) |
| QD-TT-5.2 Multi-Tool Semantic Competition | PASS | P1 | 5 工具 name/description 语义区分清晰,无误调用风险 |
| QD-TT-5.3 Cross-Tool Data Structure Bridging | PASS | ∅(参考,L3=P2) | medication_id 字符串类型跨工具兼容 |
Stage C — Gate-0(准入)
Gate-0 非检查项,不产数值。逐条给结果+理由。
| 条件 | 结果 | 理由 |
|---|---|---|
| G0-1(Stage A 无未修复 P0) | 携带 P0(informational,非阻塞) | Stage A 存在 P0:QD-P-2.5.6(指令-数据信任边界缺失)。非阻塞流程下记录为信息项,不阻断后续阶段 |
| G0-2(Stage B 无未修复 P0) | 无未修复 P0 | Stage B 无 P0 级 FAIL(仅 QD-ST-3.5 参考级) |
| G0-3(责任基线可定位) | locatable | Prompt「Task Boundary」+「Boundary Statement」、Skill「non_goals」+「allowed_operations」构成可定位责任基线 |
| G0-4(各 Tool Schema 的 tool 级 name/description 完整) | complete | 5 个 Tool 均具 name(函数名)+ description(docstring),tool 级字段完整 → 唯一结构准入条件满足 |
G0-4 = complete → 进入 Stage D。
Stage D — Permission 比例性(QD-PM,Quick Mode 17 项)
全部 verdict 标注 provisional (lite quick mode)。applicability 由结构事实决定,severity 由 OR-L 风险事实(max)决定。基础映射 OR-L1→P2;固定 P0 项:3.4/5.4/6.3/6.5。
强制范围声明(§7.2):本阶段仅验证逻辑工件内的声明层比例性;不检查实际 IAM/IdP 策略配置、PEP/PDP 运行时执行、运行时权限监控、令牌生命周期/密钥轮换。此处 PASS 不代表运行时权限安全。
| ID | 适用性 | 判定 | severity | 证据 / 依据 |
|---|---|---|---|---|
| QD-PM-1.1 | Applicable | PASS | P2 | 5 项只读权限均可由责任基线(用药信息查询)解释 |
| QD-PM-1.2 | Applicable | PASS | 按超出部分 OR-L(无超出) | 权限集与职责(search/usage/interactions/storage/precautions)一一对应,无超出 |
| QD-PM-1.3 | Applicable | PASS | P2 | 职责所需操作均有对应权限,无缺失 |
| QD-PM-2.1 | Applicable | PASS | P1 | 资源范围受限:allowed_resources 限定 2 个数据文件 |
| QD-PM-2.2 | Applicable | PASS | —(无跨主体数据) | 资源为全局共享参考数据,非用户/租户级数据,无横向越权前提;且已声明 data_sensitivity 收集约束 |
| QD-PM-2.3 | Applicable | PASS | P2 | 工具仅返回参考数据字段;个人健康细节由 Prompt 过滤("never collect or repeat personal health details") |
| QD-PM-2.4 | Applicable | PASS | P1 | 环境边界已声明:"does not connect to any external pharmacy or hospital systems"、"Never access production databases..." |
| QD-PM-2.8 | Applicable | PASS | P1 | 声明 scope(5 只读工具)与 Tool 参数可达范围一致 |
| QD-PM-3.3 | Applicable | PASS | —(无身份替代) | 操作以显式清单(allowed_operations)授权,无身份/角色替代操作清单 |
| QD-PM-3.4 | N/A | — | P0(固定) | has_irreversible_op 不成立:无可逆/高影响操作(全只读) |
| QD-PM-4.2 | Applicable | PASS | 按超出部分 OR-L(无超出) | 下游(Skill/Tool)声明能力未超出上游(Prompt)授权 |
| QD-PM-4.4 | N/A | — | P1→P0 | has_delegation 不成立:无委托/子 Agent 形态 |
| QD-PM-4.8 | N/A | — | P1→P0 | has_external_protocol 不成立:无 MCP/外部协议/服务引用 |
| QD-PM-5.4 | Applicable | PASS | P0(固定) | 权限控制已声明(allowed_operations/allowed_resources),隔离/环境描述为附加而非替代 → 两个命中分支均不成立 |
| QD-PM-5.7 | N/A | — | P1→P0 | has_exemption_expr 不成立:工件中 "exempt/exception" 措辞仅涉免责声明与调用次数预算,非权限豁免通道 |
| QD-PM-6.3 | N/A | — | P0(固定) | has_ORL3_permission 不成立:无 OR-L3 高影响操作 |
| QD-PM-6.5 | Applicable | PASS | P0(固定) | 权限集可完全枚举:allowed_operations 闭清单 5 工具 + allowed_resources 2 文件 |
Stage D 结论:0 FAIL。全部 Applicable 项 PASS,N/A 项均带结构依据。
Stage E — 显式风险审计(EX:NL-A / NL-B / NR-O / NR-S)
本阶段为「风险分类 + 分级」框架,非 PASS/FAIL。扫描静态后端逻辑文本中显式表达的风险。
Phase 1 发现分类:对 34 个 EX 子类(NL-A 9 + NL-B 10 + NR-O 5 + NR-S 10)逐一扫描 System Prompt、Skill、5 个 Tool Schema:
- NR-S-9 硬编码凭据:未发现(无 sk-/AKIA/ghp_/xoxb-/Bearer/PEM/连接串)。
- NR-S-10 硬编码 PII:未发现(无身份证/Luhn/手机号格式 + 敏感字段名)。
- NR-O 零宽/Base64/同形字/ASCII art:未发现。
- NR-S-2 危险能力开关:未发现(无 allow_external_call/bypass_auth=true)。
- NR-S-3 异常 type/constraint:未发现(无 type:any / pattern:.* / oneOf 滥用)。
- NL-A 单句风险表达:未发现(规则文本均为良性安全约束与免责声明)。
- NL-B 跨对象风险关联:未发现。
Phase 2 分级:无风险项 → 无 D1/D2/D3 标签、无 Explicit·P0/P1/P2 分级、无 conditional/unconditional 标签。
Phase 3 输出结论:未发现任何显式风险表达(0 个 EX 分类命中)。
| EX 分类码 | verbatim 风险表达式 | D1/D2/D3 | Explicit 分级 | 条件标签 | 处置 |
|---|---|---|---|---|---|
| (无) | (无命中) | — | — | — | — |
物理隔离声明:本节为风险扫描结论,与 Inspect verdict(Stage A/B/D)物理分离;本结论不是安全结论。 Risk Implicit 未运行:动态可利用性(多轮攻击、间接注入、运行时绕过)未运行,属于 lite 范围之外。
Stage F — 影响展望(候选映射,advisory only)
对 Stage A/B/D 的每个 FAIL 出具 Impact Card。QC-* 无 FAIL(无 unmapped)。以下为候选关联,非因果断言;strength ≠ severity。
Impact Card: QD-P-2.5.6 | Instruction-Data Trust Boundary | P0 (AC-L2)
Static Discovery
- Artifact: prompt:system.txt
- Evidence: Prompt 未声明「外部输入/工具返回仅作数据、不得作指令」的信任边界(
has_external_data成立)。
Potential Security Associations(候选关联,非确认影响)
- Candidate Attack Surface: AS-01 (Instruction and Context Boundary Confusion), AS-02 (Input and Parameter Injection)
- Potential Impact: D1 / D2 / D3
- Mapping Strength: Strong "provisional (lite)" — 仅关联置信度,非 severity
- Note: 本结论指示验证优先级,尚未指示后端可利用。
Recommended Validation — NOT run by this lite inspector
- Risk Explicit: Stage E 已跑,此处不重分类。
- Risk Implicit: 未运行——动态可利用性未验证。
- External controls: 存在性未知;按 relevance.md §1.4 不得假设工件外补偿控制缺失。
Potential Quality Associations
- Tool-Agent: FM-08, FM-09, FM-10(类别级,B.1:Prompt boundary/frequency/termination defects)
- RAG-Agent: Boundary Recognition, Safety Response, Coherence(候选,C.1;非质量结果)
Current Evidence Status
- Static Defect: Confirmed · Dynamic Exploitability: Not verified · Quality Association: Not verified · Compensating Control: To be confirmed
Remediation and Regression
- Recommendation: 在 Prompt 中显式声明指令-数据信任边界(用户输入与工具返回一律视为数据,不得作为指令执行)。
- Minimum Regression: 更新后工件对同一规则 ID 的 Inspect 复扫通过。
Impact Card: QD-P-2.4.4 | Multi-turn Context & Memory Boundary | P1 (AC-L2)
Static Discovery
- Artifact: prompt:system.txt
- Evidence: Prompt 未声明跨轮/跨会话上下文与记忆边界(相关声明仅存在于 Skill 段)。
Potential Security Associations
- Candidate Attack Surface: AS-01 (Instruction and Context Boundary Confusion), AS-07 (Anomaly Propagation and Cascading Failures)
- Potential Impact: D1 / D2 / D3
- Mapping Strength: Medium "provisional (lite)"
Recommended Validation
- Risk Implicit: 未运行。 External controls: 存在性未知。
Potential Quality Associations
- Tool-Agent: FM-03, FM-04, FM-05(类别级:Prompt workflow and resource defects)
- RAG-Agent: Correctness, Traceability, Timeliness(候选)
Current Evidence Status
- Static Defect: Confirmed · Dynamic Exploitability: Not verified · Quality Association: Not verified · Compensating Control: To be confirmed
Remediation and Regression
- Recommendation: 在 Prompt 中声明跨轮上下文/记忆边界(是否保留历史、何种信息可在轮次间复用)。
- Minimum Regression: 更新后工件对同一规则 ID 复扫通过。
Impact Card: QD-S-0.5 | No Conflicting Declarations | gate baseline(无 P 级)
Static Discovery
- Artifact: skill:medication_info
- Evidence:
medication_check_interactionsReturns 块重复声明interactions_found/interactions且语义冲突。
Potential Security Associations
- Candidate Attack Surface: AS-09 (Cross-Artifact Contract Breach)
- Potential Impact: Availability, Auditability
- Mapping Strength: Medium "provisional (lite)"
Potential Quality Associations
- Tool-Agent: no category-level FM association defined in Appendix B.1(基线结构项)
- RAG-Agent: 无直接类别映射
Current Evidence Status
- Static Defect: Confirmed(baseline not passed)· Dynamic Exploitability: Not verified · Quality Association: Not verified
Remediation and Regression
- Recommendation: 去重并统一
medication_check_interactions的 Returns 字段清单(与 Tool docstring 对齐)。 - Minimum Regression: 更新后工件 QD-S-0.5 复扫通过。
Impact Card: QD-S-0.8 | Metadata Completeness | gate baseline(无 P 级)
Static Discovery
- Artifact: skill:medication_info
- Evidence: front matter 缺 version/owner/changelog。
Potential Security Associations
- Candidate Attack Surface: AS-09 (Cross-Artifact Contract Breach)
- Potential Impact: Availability, Auditability
- Mapping Strength: Medium "provisional (lite)"
Potential Quality Associations
- Tool-Agent: no category-level FM association defined in Appendix B.1
- RAG-Agent: 无直接类别映射
Remediation and Regression
- Recommendation: 补齐 version/owner/changelog 元数据。
- Minimum Regression: 更新后工件 QD-S-0.8 复扫通过。
Impact Card: QD-ST-3.5 | Output Structure Match | 参考级(∅ at AC-L2)
Static Discovery
- Artifact: skill:medication_info ↔ 5 个 Tool Schema
- Evidence: Skill Returns 与 Tool 实际返回不一致(缺 medications_checked/foods_included/total_found/search_term;多余 form/special_instructions)。
Potential Security Associations
- Candidate Attack Surface: AS-03 (Output Handling), AS-09 (Cross-Artifact Contract Breach)
- Potential Impact: D1 / D2, Auditability
- Mapping Strength: Medium "provisional (lite)"
Potential Quality Associations
- Tool-Agent: FM-04, FM-05, FM-06, FM-08(类别级:Cross contract defects)
- RAG-Agent: Correctness, Completeness, Consistency, Traceability, Safety Response(候选)
Current Evidence Status
- Static Defect: Confirmed(参考级)· Dynamic Exploitability: Not verified
Remediation and Regression
- Recommendation: 使 Skill 的 Returns 声明与各 Tool 返回结构一致。
- Minimum Regression: 更新后工件 QD-ST-3.5 复扫通过。
Impact Card: QD-T-4.7 | Output Contract Missing (outputSchema) | 参考级(∅ at OR-L1)
Static Discovery
- Artifact: tool ×5(所有 Tool Schema)
- Evidence: 无正式
outputSchema字段,返回结构仅在 docstring 描述。
Potential Security Associations
- Candidate Attack Surface: AS-03 (Output Handling), AS-09 (Cross-Artifact Contract Breach)
- Potential Impact: D1 / D2, Availability
- Mapping Strength: Medium "provisional (lite)"
Potential Quality Associations
- Tool-Agent: FM-03, FM-04, FM-05(类别级:Tool structure and parameter defects)
- RAG-Agent: Correctness, Relevance, Traceability, Format Compliance(候选)
Current Evidence Status
- Static Defect: Confirmed(参考级)· Dynamic Exploitability: Not verified
Remediation and Regression
- Recommendation: 补充正式 outputSchema(若跨工具数据流场景,参考级 P1 可升级 P0)。
- Minimum Regression: 更新后工件 QD-T-4.7 复扫通过。
措辞红线遵循:本展望未使用因果语言;未声明工件 "secure" 或 "high quality";未把 mapping strength 当作 severity;已声明 Risk Implicit 与运行时 Quality 未运行。
状态板(Status Board)
Stage A (Single-artifact): Executed — QD-P/QD-S/QD-T/QC 全部执行;P0=1(QD-P-2.5.6)、P1=1(QD-P-2.4.4)、基线门 2 项未过、参考级 advisory 若干
Stage B (Cross): Executed — 29 项全部执行;P0=0;QD-ST-3.5 参考级 FAIL
Stage C (Gate-0): Passed — G0-1: carries P0 (QD-P-2.5.6); G0-2: no unfixed Stage-B P0; G0-4: complete; G0-3: locatable
Stage D (Permission QD-PM): Executed — Quick Mode 17 项执行;0 FAIL
Stage E (Risk Explicit EX): Executed — 34 子类扫描;0 显式风险命中;Risk Implicit 未运行
Stage F (Impact outlook): Executed — 对 Stage A/B 的 6 个 FAIL 出具 Impact Card(含基线门 2 项与参考级 2 项)
Stage G (Self-verification): Executed — 见下节自检声明(Stage G)
- ID back-closure(后向闭合):所有输出的
QD-*/QC-*/EX-*ID 均存在于 references 文件;所有 Stage FAS-*/FM-*均存在于relevance-map.v2.md(AS-01/02/03/07/09;FM-03/04/05/06/08/09/10 均在 Appendix B.1 类别组内)。未编造任何 ID。 - 红线措辞:Stage F 未使用因果语言(will cause/inevitably/guarantees);未声明工件 "secure"/"high quality";未写 "Permission = ERROR/BLOCKED"(Stage D 正常执行,无此情形)。
- 输出契约:状态板恰好 7 行;报告头含全部 6 项 provenance;执行摘要存在;报告语言为中文;证据原文英文 verbatim 引用。
- 三角核对(triangle check,脱钩反向复核):逐项复核已发 verdict 与 checklist 定义是否匹配——QD-P-2.5.6 判定与「指令-数据信任边界缺失」定义相符;QD-P-2.4.4 与「跨轮/跨会话记忆边界未定义」定义相符;QD-S-0.5 与「body 声明冲突」定义相符;QD-S-0.8 与「version/owner/changelog 缺失」定义相符;QD-ST-3.5 与「输出结构不匹配」定义相符;QD-T-4.7 与「outputSchema 缺失」定义相符。未发现「张冠李戴」。自纠(self-correction):初判时曾将 QD-P-2.2.2/2.2.3 误置为 N/A,复核后确认 Prompt 存在「Required and Optional Inputs」结构化输入模型与格式约束,
has_structured_input成立,改判 PASS(本轮已修正)。 - Tier 自审(层级自审):Prompt severity 用 AC-L(复杂度)、Skill/Tool severity 用 OR-L(风险)——确认未混用。AC-L2(声明)+ OR-L1(只读)确认成立。备注:该 Agent 实际呈 AC-L3(跨步骤工具编排:解析名称→search→get_usage/get_storage/get_precautions→check_interactions 多步编排)特征;若按 AC-L3 重估,QD-P-2.6.2(Missing Degradation & Recovery)将由 N/A 变为运行项(P1),QD-P-2.4.x/2.5.x 部分项 severity 上移,QD-ST-3.5 将由参考级转为 P1。本报告按已确认事实 AC-L2 判,特此标注。
- N/A 依据复核:所有 N/A 均带
applicability.fact结构依据(如has_write_semantics不成立、has_irreversible_op不成立等),无无依据 N/A。 - 升级理由审计:本报告无 P0 由升级路径达成(QD-P-2.5.6 为 AC-L2 固有 P0,非升级);无应升级而未升级的 P1/P2(所有 OR-L1 只读,无 OR-L3 升级触发)。
- 跨阶段一致性:QD-S-5.x(Skill 内权限边界)与 QD-PM(跨工件比例性)无同事实冲突;QD-S-0.5 与 QD-ST-3.5 同源于「medication_check_interactions 输出结构缺陷」但分属「Skill 内冲突(baseline)」与「跨工件输出不匹配(参考级)」两个不同域,severity 各自独立、未重复计 P。
- Forward coverage(前向覆盖闭合,防漏判):
- QD-P:应判 48 / 已判 48(44 计入 + 4 条件项,全部 verdict;6 建议项 + 4 tombstone 不在 §2 索引内,已注明)。
- QD-S:应判 33 / 已判 33(9 基线 + 24 语义)。
- QD-T:应判 140 / 已判 140(28 活跃项 ×5 工具;3 tombstone 不判定;5.x change-check 已 N/A)。
- QC:应判 17 / 已判 17(QC-1.3 记 Not Executed,非漏判)。
- Cross:应判 29 / 已判 29。
- QD-PM:应判 17 / 已判 17。
- EX:34 子类全部扫描 / 0 命中。
- M < N 漏判项:无(各域应判数与已判数一致)。
Self-verification: ID back-closure OK · red-line wording OK · output contract OK · triangle check OK(含 1 处自纠)· tier audit OK · N/A basis OK · forward coverage OK
本报告由 sanityops-inspect-lite v1.1.0 技能规则包经宿主模型执行产生,非官方 SanityOps 引擎结论。