一键导入
result-verification
Use before marking any workflow stage as completed — verify outputs exist, are valid, and meet success criteria defined in workflow-state.json
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use before marking any workflow stage as completed — verify outputs exist, are valid, and meet success criteria defined in workflow-state.json
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when creating, reading, or updating workflow checkpoint files (workflow-state.json, progress.log, init.sh) — the file-centric state system for cross-session computational workflows
Use when a workflow stage subagent has prepared computation scripts and needs to submit the job to a local or HPC backend via DPDispatcher
Use when a workflow plan exists and stages need to be executed sequentially in the current session
Use when the user wants to set up a new computational experiment — before any computation begins, to design the method, stages, and success criteria
Use when starting a new session and workflow-state.json exists in the working directory — recovers full context from checkpoint files before any new work
Use when any workflow stage fails — simulation, analysis, plotting, post-processing, or environment setup — before proposing fixes
| name | result-verification |
| description | Use before marking any workflow stage as completed — verify outputs exist, are valid, and meet success criteria defined in workflow-state.json |
Claiming a stage is complete without verification is dishonesty, not efficiency.
Core principle: Evidence before claims, always.
Violating the letter of this rule is violating the spirit of this rule.
NO STAGE MARKED COMPLETED WITHOUT PASSING VERIFICATION
If verification fails, the stage is marked failed and superscientist:systematic-debugging is invoked.
You MUST complete every check. A single failure = stage is failed.
# For each expected output in workflow-state.json stage.outputs:
test -s "$output_file" || echo "FAIL: $output_file missing or empty"
Read the success_criteria from workflow-state.json for this stage. Check each criterion against actual output:
| Criterion type | How to verify |
|---|---|
| Convergence metric | Parse output, check final value meets threshold from success_criteria |
| Numerical result | Extract value, compare against target or reference |
| Output file produced | Check file exists, non-empty, and in expected format |
| Visualization | Check image file valid, non-zero size, expected content present |
| Script/process completion | Check exit code is 0 and no fatal errors in logs |
Domain-specific reasonableness checks (derive from experiment design):
Check log/output files for:
After verification, report:
If passed:
Verification PASSED for stage-N ([name]):
- All N output files present and non-empty
- Success criteria met: [specific evidence]
- Sanity checks passed: [brief summary]
→ Marking stage as completed.
If failed:
Verification FAILED for stage-N ([name]):
- Failed check: [which check failed]
- Evidence: [what was found]
- Expected: [what was expected]
→ Marking stage as failed. Invoking systematic-debugging.
| Thought | Reality |
|---|---|
| "Output file exists, good enough" | Check it's non-empty and parseable too. |
| "Calculation converged, must be right" | Check the answer is physically reasonable. |
| "No errors in the log" | Check for warnings too. |
| "I'll verify the next stage instead" | Every stage gets verified. No shortcuts. |
| "Results look reasonable" | Compare against success criteria, not intuition. |