| name | full-qa-audit |
| display_name | 原始记录全量质检 |
| description | 对原始记录执行全流程质检,包含要素合规性、逻辑校验与数据合理性检查。 |
Skill: full-qa-audit
Description
This is a Master Orchestrator Skill designed to execute a complete Quality Assurance Pipeline on inspection projects. It coordinates three specialized sub-skills (record-amendment-audit, raw-record-vs-report-consistency, report-audit) to verify compliance, consistency, and correctness.
Input Contract
- Raw Records: One or more files (Images/PDFs) containing original inspection data.
- Report: A final inspection/appraisal report document.
Execution Workflow (Pipeline)
You must strictly follow this sequential process. Do not skip steps unless a "Stop Condition" is met.
🛑 Step 1: Raw Record Sanity Check (Foundation)
- Action: Invoke the logic of
record-amendment-audit.
- Focus: Check for unassigned amendments, missing signatures, or illegible data.
- Decision Point:
- If
overall_assessment is "C (Major Non-compliance)" or worse: -> TERMINATE PIPELINE.
- Synthesize the findings into the "step_1_result" section of the output.
- Else: Proceed to Step 2.
🔗 Step 2: Consistency Cross-Check (Linkage)
- Action: Invoke the logic of
raw-record-vs-report-consistency.
- Focus: Verify if Dates, Instrument IDs, and Record IDs match between the valid Raw Records (from Step 1) and the Report.
- Decision Point:
- If
overall_status is "fail" (e.g., mismatched instruments): -> TERMINATE PIPELINE (No need to audit a report based on wrong data).
- Synthesize the findings into the "step_2_result" section of the output.
- Else: Proceed to Step 3.
📝 Step 3: Report & Conclusion Audit (Final Product)
- Action: Invoke the logic of
report-audit.
- Focus: Check the report's internal consistency, typos, standard compliance, and logic.
- Final Actions:
- Synthesize the findings into the "step_3_result" section.
- Generate the final
overall_pipeline_conclusion.
Output Structure
Return a single consolidated JSON object. See output_schema.json.
Best Practices
- Context Management: When passing context to Step 2, prioritize the dates and IDs extracted in Step 1 to save effort.
- Fail Fast: Do not waste resources auditing the grammar of a report if the raw data is invalid.