lov-skill-optimizer

name	lov-skill-optimizer
category	Meta Skills
tagline	Audit an Agent Skill, apply focused fixes, bump semver, and verify every distribution layer.
description	Audit and optimize one or more existing Agent Skills from a canonical source path, then bump semver, update README/SKILL.md/skill.yaml/CHANGELOG.md, and verify installed copies and catalog synchronization. Checks frontmatter, trigger quality, CLI hygiene, naming, portability, version drift, dirty worktrees, shared Skill feedback policy, and compatibility contracts. Use when the user asks to optimize, refine, audit, polish, or update a Skill, or mentions "优化 skill", "skill 审计", "刷一遍 skill", "skill-optimizer", or "update skill changelog".
license	MIT
compatibility	Requires Python 3.8+ (stdlib only). Git is optional for linting and required for source commit/push verification. Catalog synchronization is discovered from explicit paths, environment variables, or nearby checkouts; unavailable locations are reported rather than assumed.
metadata	{"author":"lovstudio","version":"0.10.0","tags":"meta skill-maintenance versioning changelog lint portability sync"}

skill-optimizer — 自动优化 Agent Skill 并维护版本与分发状态

This is a non-interactive maintenance workflow. It infers the target and prioritizes issues from the current conversation, then supplements them with a generic lint pass. When several Skills are named in one request, process them in the order named and emit a separate result block for each Skill.

Target and source resolution

Prefer an explicit canonical path whenever the Skill is outside a conventional skills repository:

python3 scripts/lint_skill.py --path /absolute/path/to/skill --json
python3 scripts/inspect_layout.py --path /absolute/path/to/skill --json

For a name, accept foo, lov-foo, or foo-skill. Resolution may find a source checkout, an installed copy, or a catalog entry. Before editing:

1. Resolve symlinks and record the actual path.
2. Identify the Git root, branch, and dirty worktree state.
3. Treat a source checkout as canonical. If the target is an installed copy, locate a matching source checkout; when no source exists, report that the supplied path itself is canonical instead of silently editing another copy.
4. Record every discovered installation and catalog path. A copy is synced only when its content digest matches the source; a missing location is not_discovered, never complete. A symlink is synced only when it resolves to the canonical source.

Do not absorb pre-existing edits into a maintenance commit. If a target file is already dirty, review the overlap before editing it and stage only the exact files changed by this optimization.

Workflow (mandatory)

Step 0: Classify feedback scope

When the optimization is triggered by a user correction, classify it before choosing targets:

• task-specific: applies only to the current artifact or one-off value. Do not edit a Skill.
• skill-specific reusable: applies to future runs of one domain or platform. Optimize the relevant canonical Skill.
• global reusable: changes how every Skill should handle feedback, authorization, sequencing, or handoff. Update the host's active user-level shared instruction artifact once (for example the applicable user-level AGENTS.md); do not paste the same policy into every domain SKILL.md.

If a global policy was previously placed in one domain Skill, move it to the shared layer and remove the domain duplicate while preserving any genuinely domain-specific rule learned in the same incident. A shared policy file may be outside the Skill repository; report its path and verification state explicitly instead of pretending it is part of the Skill package.