Skill Lifecycle Manager

Manage skill work as a lifecycle, not as isolated one-off edits. This skill is a thin router: it figures out which lifecycle stage matters now, runs only those stages, and hands off deep specialist work to the right sub-skill or helper instead of re-embedding every workflow inline.

The default design bias is:

• Use OpenAI-style structure discipline for authoring, packaging boundaries, and progressive disclosure
• Use Anthropic-style evaluation discipline for realistic test prompts, baseline comparisons when warranted, and separate trigger optimization
• Use repo-local conventions for skill.json, CHANGELOG.md, categories, generated catalogs, and validation commands

Outputs

Depending on the requested stage, produce one or more of:

• An invocation posture decision for the target skill: manual-first, hybrid, or auto-first, with rationale
• A new or updated skill folder with SKILL.md, skill.json, CHANGELOG.md, and any required references/scripts/assets
• A fusion brief or source inventory explaining what to borrow from existing skills
• A validation report with concrete failures and fixes
• An evaluation plan or iteration workspace path under workspaces/<skill-name>/
• A seeded evaluation workspace with with-skill and optional baseline slots