cultivate-workflowlisted

# Cultivate Workflow Preserve a proven scenario as one scenario Skill plus the smallest readable Skill sequence. Do not create a second orchestration language. ## Before Writing 1. Identify the concrete repeated Goal and inspect the successful Cases, explicit feedback, responsible Skills, Artifact sequence, and real exceptions. One user request can justify cultivation, but speculative flexibility cannot. 2. Confirm every stable step is an adopted local Skill under `skills/`. A Candidate is not callable; a Trial may serve its current Case but must be explicitly adopted before entering the stable Workflow. 3. Reuse each Skill's existing `## 完成标准`. Do not copy its business method, Prompt, scripts, templates, or quality obligations into the Workflow. ## Scenario Package Create only: ```text workflows/<scenario-id>/ ├── SKILL.md ├── workflow.yaml └── references/ └── benchmark.md # only when the Benchmark is too large for SKILL.md ``` The scenario `SKILL.md` defines when the Goal matches, which Profile or Case inputs matter, how the current Host uses the Workflow, the final outcome Benchmark, stopping conditions, and `## 完成标准`. Its description must let Codex App or Claude Code select it from a short Goal. The Benchmark judges the final user-visible outcome. Each criterion states: - what acceptable means; - what Artifact, screenshot, or objective signal proves it; - which Owner Skill must be rerun when it fails; - how to proceed honestly when the evidence or actio