tywinlu1988/Credence-China / dev/.claude/skills/credit-qa-verifier
credit-qa-verifier
VerifiedUse when verifying a Chinese fixed-income credit report or analysis — checking a work path's quality gates, enforcing the mandatory signal-density rules (no numeric score below the density floor), the one-shot-veto CCC ceiling, Mode B anti-hallucination guardrails, and single-source-of-truth compliance (no invented thresholds). Triggers on '质检', '复核这份报告', '检查一下有没有问题', '质量门', or as the final step after report generation. Reads engine documents as the rule source; never relaxes a gate.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 7.8 KB |
| references/qa-checklist.md | markdown | 2.8 KB |
SKILL.md
View on GitHub---
name: credit-qa-verifier
description: Use when verifying a Chinese fixed-income credit report or analysis — checking a work path's quality gates, enforcing the mandatory signal-density rules (no numeric score below the density floor), the one-shot-veto CCC ceiling, Mode B anti-hallucination guardrails, and single-source-of-truth compliance (no invented thresholds). Triggers on '质检', '复核这份报告', '检查一下有没有问题', '质量门', or as the final step after report generation. Reads engine documents as the rule source; never relaxes a gate.
---
## Purpose
**对应引擎版本**: v0.12.3-release
质检层,四段链**终态**。职责:对《交付单》及其上游《分析产物》《工作路径单》做交付前复核,产出《质检裁决》(QA Verdict)。本 skill 以引擎文档为规则源,**从不放宽门禁**(never relaxes a gate):任何一项质量门或强制检查不通过,即判 `fail` 并退回整改,不得为了交付而降低标准。本 skill 不复制任何阈值/评级映射;规则正文以所引引擎文档为单一事实源。
## Inputs(消费)
- **《交付单》(Delivery Note)**:`credit-report-builder` 产出(字段形状见 `dev/engine/pipeline-contract.md` §2.3)。
- **《分析产物》(Analysis Artifact)**:`fixed-income-credit-analysis` 产出,供复核密度/否决/完备性(§2.2)。
- **《工作路径单》的 `quality_gates`**:`credit-analysis-router` 产出,是逐门复核的清单来源;`path_id` 为贯穿 join key,三份产物须一致。
## Output(产出)
- **《质检裁决》(QA Verdict)**:终态产物,字段形状见 `dev/engine/pipeline-contract.md` §2.4。`verdict` 取值 `pass` | `pass-with-findings` | `fail`。
## Verification Protocol(复核协议)
1. **join key 一致性**:三份产物的 `path_id` 必须相同且在注册表可解析;不一致即 `fail`。
2. **逐质量门复核**:对路径单 `quality_gates` 逐条复核,产 `gate_results`(每门 `status` + `evidence`,证据引用引擎文档章节)。
3. **八项强制检查**:见下「Mandatory Checks」,任一不通过即 `fail`。
4. **产裁决**:全部通过 → `pass`;通过但有应注记的发现 → `pass-with-findings`;任一不通过 → `fail` 并列 `remediation`。
## Mandatory Checks(强制检查,规则源为引擎文档)
- **信号密度规则 `density_rule`**:密度低于下限的维度不得出数值评分,须标注"信息不足无法评估";加权密度不足时不得出最终字母评级。规则源 `dev/engine/mosaic-engine.md` §4.3。
- **一票否决上限 `veto_ceiling`**:触发一票否决的发行人,评级上限锁定为 CCC 不得上调。规则源 `dev/engine/industry-framework.md` §五。
- **Mode B 防幻觉 `mode_b`**:用户未显式提供数据源(CSV/API/MCP)时,不得出现任何 Mode B 外部数据值;所有 Mode B 字段须作数据缺口处理。规则源 `dev/engine/mosaic-engine.md` §六。
- **单一事实源 `single_source`**:报告/分析不得编造阈值、权重、评级映射;引擎未定义的量须标注 `引擎未定义`。规则源为全部所引引擎文档。
- **模板一致性 `template_fidelity`**:交付报告的结构(章节序列与组件)必须与该路径 registry `templates` 字段指定的模板逐节对应;自行设计报告版式、"参考模板风格自制"、模板结构外自创章节,一律 `fail`。规则源 `dev/engine/pipeline-contract.md` §五。
- **方法论一致性 `methodology_fidelity`**:分析产物与报告中的分析维度、评分体系、分析框架必须可溯源至引擎文档具体章节;自造维度/框架、以通用信用分析先验补位,一律 `fail`。规则源 `dev/engine/pipeline-contract.md` §五。
- **文件交付 `file_delivered`**:`rendered` 列出的每个报告文件必须实际存在于本地文件系统且可读取(文件存在性验证);文件缺失 = `fail`。若 `rendered` 报告数 >2 且缺 `report-index.html` → `pass-with-findings` 并提醒补索引页。
- **CSS 独立性 `css_independent`**:交付的每个 HTML 报告文件不得包含外部样式引用(`<link rel="stylesheet"` 或 `href=".css"` 任一命中即 `fail`);CSS 须为内联 `<style>…</style>` 形态。规则源 `dev/engine/pipeline-contract.md` §五。
## QA Verdict Output(《质检裁决》)
模板(schema 单一事实源为 `dev/engine/pipeline-contract.md` §2.4):
```yaml
path_id: "" # join key(三份产物须一致)
verdict: "" # pass|pass-with-findings|fail
gate_results: # 逐质量门复核结果
- gate: "" # "规则名 (dev/engine/<doc>.md §节)"(承自路径单 quality_gates)
status: "" # pass|fail
evidence: "" # 复核证据(引用引擎文档章节,不复制数值)
mandatory_checks: # 八项强制检查
density_rule: ""
veto_ceiling: ""
mode_b: ""
single_source: ""
template_fidelity: ""
methodology_fidelity: ""
file_delivered: ""
css_independent: ""
remediation: [] # 不通过项的整改建议
```
示例(审贷单标的 WP-M0-01,三门复核通过、强制检查全过,有一项应注记发现):
```yaml
path_id: WP-M0-01
verdict: pass-with-findings
gate_results:
- gate: "信号密度 (dev/engine/mosaic-engine.md §4.3)"
status: pass
evidence: 低于密度下限的维度已置 null 并标注信息不足无法评估(mosaic-engine §4.3)
- gate: "一票否决 (dev/engine/industry-framework.md §五)"
status: pass
evidence: 未触发一票否决,评级上限规则复核通过(industry-framework §五)
- gate: "交叉对撞 (dev/engine/dual-track-methodology.md §四)"
status: pass
evidence: 双轨分歧已在报告中呈现并解读(dual-track-methodology §四)
mandatory_checks:
density_rule: pass
veto_ceiling: pass
mode_b: pass
single_source: pass
template_fidelity: pass
methodology_fidelity: pass
file_delivered: pass
css_independent: pass
remediation: []
```
## Chaining(链式交接 · 终态)
- **上游(REQUIRED)**:`credit-report-builder` —— 消费其《交付单》及上游产物。本 skill 由上游**自动触发**,无需用户指令;不得就"是否需要质检"询问用户。
- **终态**:本 skill 为四段链最后一段,无下游。判 `pass` / `pass-with-findings` 时直接回执裁决并结束,不问用户。判 `fail` 时先按 `remediation` **自动**退回对应阶段(密度/否决问题回 analysis;模板/装配问题回 report)整改一轮后重新质检(自愈);仅当一轮自愈后仍 `fail`,才向用户出示裁决与整改选项。全链交互点预算见 `dev/engine/pipeline-contract.md` §三「链式接续规则」。
## Guardrails
- **从不放宽门禁(never relaxes a gate)**:质量门与强制检查只有"通过/不通过",不设"酌情通过"。缺完备性报告、低密度出数值分、编造阈值、Mode B 幻觉,一律 `fail`。
- **不复制引擎内容**:只引用规则名与文档章节,不复制任何阈值、SRI 档位、分层时间预算或评级映射;数值裁决以所引引擎文档为准。
- **以引擎文档为规则源**:每条 gate 规则名必须能在所引引擎文档中 grep 到(溯源见 `references/qa-checklist.md`),不得虚构规则。
## References
- `references/qa-checklist.md` — 质检清单(逐门规则名 → 引擎文档溯源,单源指针)
- `dev/engine/pipeline-contract.md` — 四段链 I/O 契约(产物 schema 单一事实源)
- `dev/engine/mosaic-engine.md` — 信号密度/完备性/Mode B 护栏(density_rule、mode_b 规则源)
- `dev/engine/industry-framework.md` — 一票否决评级上限(veto_ceiling 规则源)
- `dev/engine/work-path-registry.md` — 工作路径注册表(质量门清单溯源)
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 323k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |