Skill Optimizer Plus

Problem

Existing SKILL.md files often mix capability descriptions, trigger conditions, workflow details, and platform assumptions. This can lead to misfiring, overly long bodies, dangling resource references, happy-path-only SOPs, and weak fallback or rollback constraints when external dependencies fail.

How It Works

Diagnosis mode is the default and does not rewrite the target skill. It produces an optimization report and an action list:
- validates YAML frontmatter, name, description, directory layout, and unnecessary files;
- checks whether description states what the skill does, when to use it, keyword coverage, and negative boundaries;
- reviews imperative writing style, length, and whether detailed material is split into references/;
- audits MCP or service-dependent SOPs with a 5D framework for declared dependencies, degraded paths, failure isolation, endpoint state, and documentation layering.

Meta-upgrade mode runs only after the user confirms the target skill, strategy, and change scope. It uses governance_guard.py for pre-write checks, skill_interface.py for backups and deterministic edits, and coupling_analyzer.py for resource-contract re-checks. Errors such as HookDenied or ValidationError are recorded and rolled back. It can also add history.md and memory.md scaffolding based on policy, without automatically rewriting the target logic.

Boundaries

It fits skill review and governance, especially trigger quality, cross-platform compatibility, resource directories, and SOP robustness. It is not a substitute for product decisions and cannot bypass hard governance blocks for name changes, file deletion, or cross-skill writes.