Compare visible Variant A and Variant B outputs, fill the decision file, then run adjudication. The answer key is intentionally hidden from this page.
+Turn this repeated workflow into a reusable team skill package.
+Output names the routeable Skill entrypoint.
1.0Output includes neutral interface metadata.
1.0Output points reviewer to generated overview report.
1.0Output preserves entrypoint/resource separation.
1.0Create a routeable SKILL.md, agents/interface.yaml, reports/skill-overview.html, and a focused eval plan. Keep the root entrypoint lean, move durable guidance into references/, scripts into scripts/, and evidence into reports/.
I can write a prompt for that workflow and include a short checklist.
Upgrade this production skill so we know whether the generated output is better than baseline.
+Output explicitly compares with-skill and baseline outputs.
1.0Output includes objective assertion grading.
1.0Output produces a scorecard report path.
1.0Output records failure taxonomy.
1.0Add Output Eval Lab cases with baseline_output and with_skill_output, run assertion grading, report with-skill vs baseline pass-rate delta, and record failure taxonomy plus next fixes in reports/output_quality_scorecard.md.
Add more examples and run the trigger eval again.
Package this skill for OpenAI, Claude, Agent Skills, and generic targets.
+Output requires Skill IR before packaging.
1.0Output lists core IR contract fields.
1.0Output names requested runtime targets.
1.0Output says targets must preserve the capability contract.
1.0Export folders for each platform and adjust files as needed.
Export Skill IR first with job_to_be_done, trigger_surface, workflow, resources, eval_plan, risk, and governance. Then compile or package targets from the IR so OpenAI, Claude, Agent Skills, and generic packages preserve the same capability contract.
I only need a one-off summary of these notes, not a reusable process.
+Output refuses unnecessary skill creation for one-off work.
1.0Output labels the request as near-neighbor instead of owned work.
1.0Output asks for repeat-use evidence before packaging.
1.0Create a SKILL.md and a reusable workflow anyway so future notes can use it.
Do not create a skill for this one-off request. Treat it as a near-neighbor: answer the summary directly unless the user confirms repeated use, shared ownership, or a reusable output contract.
Turn the attached release brief source into a governed skill package.
+Output names file-backed source evidence.
1.0Output preserves governed ownership metadata.
1.0Output preserves output and rollback boundaries.
1.0Output requires trust and output scorecard artifacts.
1.0Output forbids invented launch evidence.
1.0Draft a release announcement with the changelog, support notes, and owner name.
Use the file-backed fixture as source evidence, then create a governed skill package with SKILL.md, agents/interface.yaml, owner, review cadence, input_files, output contract, rollback boundary, trust report, and reports/output_quality_scorecard.md. Mark missing launch metrics as missing evidence instead of inventing them.
Use this shape in reports/output_review_decisions.json; leave a case blank when the reviewer is not ready.
{
+ "schema_version": "1.0",
+ "reviewer": "",
+ "reviewed_at": "",
+ "decisions": [
+ {
+ "case_id": "skill-package-contract",
+ "winner_variant": "",
+ "confidence": null,
+ "reason": ""
+ },
+ {
+ "case_id": "output-eval-expectation",
+ "winner_variant": "",
+ "confidence": null,
+ "reason": ""
+ },
+ {
+ "case_id": "ir-before-packaging",
+ "winner_variant": "",
+ "confidence": null,
+ "reason": ""
+ },
+ {
+ "case_id": "near-neighbor-boundary",
+ "winner_variant": "",
+ "confidence": null,
+ "reason": ""
+ },
+ {
+ "case_id": "file-backed-governed-package",
+ "winner_variant": "",
+ "confidence": null,
+ "reason": ""
+ }
+ ]
+}5 targets in platform-neutral contract
target contracts compiled from Skill IR
5 cases; 1 file-backed
command 10; model 0; recorded 0
review pairs hide baseline vs with-skill labels
pending 5; answer key hidden
adjudication decisions; pending 5
2.0 coverage; evidence pending 4
target conformance pass rate
0 native; 4 installer-enforced
93 scripts scanned; secrets found
154 files scanned for Python 3.11
899 largest lines; 34 CLI handlers
12 scanned skills; route collisions
1 metadata events; 0 missed triggers
0 gates covered; human risk decisions
0 valid submissions; 0 invalid
70 public surfaces scanned
0 open blocker annotations
5 targets; MIT license
567 zip entries; package verification
4 adapters; 12 permissions enforced; 0 permission failures
declared minor; 0 breaking changes
5 targets in platform-neutral contract
target contracts compiled from Skill IR
5 cases; 1 file-backed
command 10; model 0; recorded 0
review pairs hide baseline vs with-skill labels
pending 5; answer key hidden
adjudication decisions; pending 5
2.0 coverage; evidence pending 4
target conformance pass rate
0 native; 4 installer-enforced
93 scripts scanned; secrets found
154 files scanned for Python 3.11
899 largest lines; 34 CLI handlers
12 scanned skills; route collisions
1 metadata events; 0 missed triggers
0 gates covered; human risk decisions
0 valid submissions; 0 invalid
71 public surfaces scanned
0 open blocker annotations
5 targets; MIT license
568 zip entries; package verification
4 adapters; 12 permissions enforced; 0 permission failures
declared minor; 0 breaking changes
intent confidence 100/100; Intent is clear enough to package the first routeable version.
13 trigger cases; 0 misroutes; 0 ambiguous
5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5
initial load 944/1000; quality density 137.7
5 / 5 targets pass
0 secrets; 93 scripts; 3 network-capable scripts; 0 help smoke failures
Python 3.11; 154 files; 0 compatibility issues; 0 syntax; 0 f-string 3.11 hazards
151 Python files; 0 hotspots; 0 blockers; largest 899 lines; 34 CLI handlers
3/3 permissions approved; gaps 0; required file_write, network, subprocess
4/4 targets probed; native 0; metadata fallback 4; installer 4; residual risks 4
12 skills, 1 actionable; 0 actionable route collisions; 0 actionable owner gaps; 0 actionable stale; 0 actionable drift; 24 scoped non-actionable issues
1 metadata events; adoption 0; missed 0; bad-output 0; risk low
0 active waivers; 1 warning gates still need reviewer decision
4 pending world-class evidence entries; 1 human pending; 3 external pending; overclaim guard true
yao-meta-skill 1.1.0; 6/6 compatibility entries pass; install pass with 4 adapters; installer permissions 12 enforced / 0 failures
0 promote; 3 keep current; 0 blocked; upgrade minor declared / minor recommended
intent confidence 100/100; Intent is clear enough to package the first routeable version.
13 trigger cases; 0 misroutes; 0 ambiguous
5/5 cases; with-skill 100.0; baseline 0.0; file-backed 1; near-neighbor 1; blind A/B 5; exec 10; command 10; model 0; recorded 0; reviewed 0/5; review pending 5
initial load 944/1000; quality density 137.7
5 / 5 targets pass
0 secrets; 93 scripts; 3 network-capable scripts; 0 help smoke failures
Python 3.11; 154 files; 0 compatibility issues; 0 syntax; 0 f-string 3.11 hazards
151 Python files; 0 hotspots; 0 blockers; largest 899 lines; 34 CLI handlers
3/3 permissions approved; gaps 0; required file_write, network, subprocess
4/4 targets probed; native 0; metadata fallback 4; installer 4; residual risks 4
12 skills, 1 actionable; 0 actionable route collisions; 0 actionable owner gaps; 0 actionable stale; 0 actionable drift; 24 scoped non-actionable issues
1 metadata events; adoption 100.0; missed 0; bad-output 0; risk low
0 active waivers; 1 warning gates still need reviewer decision
4 pending world-class evidence entries; 1 human pending; 3 external pending; overclaim guard true
yao-meta-skill 1.1.0; 6/6 compatibility entries pass; install pass with 4 adapters; installer permissions 12 enforced / 0 failures
0 promote; 3 keep current; 0 blocked; upgrade minor declared / minor recommended
补足 output eval 覆盖、execution evidence、blind A/B 和 reviewer adjudication。
没有输出质量和人工盲评证据时，Skill 只能证明会触发，不能证明输出真的更好且经得起审查。python3 scripts/adjudicate_output_review.py --write-template && python3 scripts/yao.py output-review对保留的 warning 写入 reviewer、理由、范围和到期时间，或修掉 warning。
warning 可以被接受，但必须可审计、会过期，并且不能掩盖 blocker。python3 scripts/render_review_waivers.py .补齐 provider、真人盲评、原生权限执行和真实客户端遥测证据，或明确本次发布不声明 world-class 完成。
世界级结论必须来自已接受的外部/人工证据；计划、metadata fallback、待评审和本地命令都不能替代完成证据。python3 scripts/yao.py world-class-submission-review . && python3 scripts/yao.py world-class-ledger . && python3 scripts/yao.py review-studio .补足 output eval 覆盖、execution evidence、blind A/B 和 reviewer adjudication。
没有输出质量和人工盲评证据时，Skill 只能证明会触发，不能证明输出真的更好且经得起审查。python3 scripts/adjudicate_output_review.py --write-template && python3 scripts/yao.py output-review对保留的 warning 写入 reviewer、理由、范围和到期时间，或修掉 warning。
warning 可以被接受，但必须可审计、会过期，并且不能掩盖 blocker。python3 scripts/render_review_waivers.py .补齐 provider、真人盲评、原生权限执行和真实客户端遥测证据，或明确本次发布不声明 world-class 完成。
世界级结论必须来自已接受的外部/人工证据；计划、metadata fallback、待评审和本地命令都不能替代完成证据。python3 scripts/yao.py world-class-submission-review . && python3 scripts/yao.py world-class-ledger . && python3 scripts/yao.py review-studio .先看 reports/output_blind_review_pack.md 做盲评，填入 reports/output_review_decisions.json，再用 reports/output_review_adjudication.md 核对答案 key。
先打开 reports/output_review_kit.html 做盲评，填入 reports/output_review_decisions.json，再用 reports/output_review_adjudication.md 核对答案 key。
reports/output_execution_runs.md 会标明 recorded fixture、command run 或 model run；只有 provider runner 返回 model metadata 时才算 model-executed。
09013a0df3555b5f1ef58254c1bc78ebd2383f8efbfe67a7a9c1010a172fb4925b6eb1631e77feb6fe3f4199e128607cdc719f8ec0675c92ae986aad2553a045高风险 secret、远程 inline execution、缺失依赖策略或无法解释的脚本接口应阻断 governed release。
1 metadata events; adoption 0; missed 0; bad-output 0; risk low
1 metadata events; adoption 100.0; missed 0; bad-output 0; risk low
这里列出每个 world-class 证据项的当前状态、完成定义、证据来源、隐私约束和下一步；计划、metadata fallback、待评审和本地命令不会被当成完成证据。
-Collect at least one provider-backed output-eval holdout run with model, timing, and token metadata.
Record real blind A/B reviewer decisions before claiming human output review completion.
Prove at least one real target client or external installer runtime guard enforces approved high-permission capabilities.
Import production metadata-only events from a real external client into the local drift loop.
Collect at least one provider-backed output-eval holdout run with model, timing, and token metadata.
Record real blind A/B reviewer decisions before claiming human output review completion.
Prove at least one real target client or external installer runtime guard enforces approved high-permission capabilities.
Import production metadata-only events from a real external client into the local drift loop.
claim guard 扫描 README、docs 和 reports 中的完成态表述；ledger 未 ready 时，任何英文完成断言、true 状态声明或中文完成态都会阻断发布审查。
yao-meta-skill 1.1.0; 6/6 compatibility entries pass; install pass with 4 adapters; installer permissions 12 enforced / 0 failures
2fbe2e4a06aa71dacd59865cf34adb163940972d04316ebbeb479a83cb46bd34fc334a4b4da4af6c463510a76551e018633deb5626dca38d9aaaa2c636591eb70 promote; 3 keep current; 0 blocked; upgrade minor declared / minor recommended
2fbe2e4a06aa71dacd59865cf34adb163940972d04316ebbeb479a83cb46bd34fc334a4b4da4af6c463510a76551e018633deb5626dca38d9aaaa2c636591eb7让 reviewer 快速确认关键文件、目录和资产分布。Lets reviewers confirm key files, directories, and asset distribution quickly.
| 路径Path | 作用Role | 类型Type |
|---|---|---|
| SKILL.md | Skill 入口文件Skill entrypoint | 文件file |
| README.md | 人类可读使用说明Human-readable usage guide | 文件file |
| agents/interface.yaml | 跨平台接口元数据Neutral interface metadata | 文件file |
| manifest.json | 生命周期与打包元数据Lifecycle and portability metadata | 文件file |
| references | 扩展指导与复用资料Extended guidance and reusable notes | 目录folder |
| scripts | 确定性脚本或本地工具Deterministic helpers or local tooling | 目录folder |
| evals | 触发与质量检查Trigger and quality checks | 目录folder |
| reports | 生成的证据与总结报告Generated evidence and overview artifacts | 目录folder |