enhance

# Enhance: certainty-graded surface correction Run a `correct` op over AI-facing project surfaces. The invariant: plugin manifests, agents, skills, prompts, docs, command definitions, hooks, and project memory must be explicit, bounded, and internally consistent. Native shape: no external analyzer binary, no persistent suppression learning, no model routing. Use ODIN discovery tools plus parallel generic `task` agents. Each analyzer emits structured findings with `certainty`, `autoFix`, `file`, `line`, `check`, `evidence`, `fix`. The orchestrator aggregates, deduplicates, and reports. ## When to Apply / NOT Apply when the user asks to enhance, improve, audit, or tighten plugin config, agent definitions, skills, CLAUDE.md/AGENTS.md, docs, prompts, commands, or hooks. NOT when the user asks for general code assessment, runtime bug fixing, visual design, performance profiling, or content rewriting without configuration/agent-surface analysis. ## Flags - `--apply`: apply only HIGH-certainty findings with `autoFix: yes`. No flag → report only. - `--verbose`: include MEDIUM and LOW findings. Default report shows HIGH plus summary counts for MEDIUM/LOW. - `--focus=<type>`: optional narrowing to one or more comma-separated analyzer families: `plugin`, `agent`, `skill`, `docs`, `prompt`, `claudemd`, `hooks`, `cross-file`. Deleted on purpose: auto-suppression learning, reset/export suppression flags, editor-platform adapters, model pins, binary cache state. ## Discovery Resol...