An agent skill that audits iOS/macOS Swift codebases for common autopilot failure modes, scores control loss, and writes a reclaim plan. Works with coding agents that support skills - not tied to a single app. It stays read-only; remediation is left to you (or a sibling fix skill) when you ask.
Open source · MIT · Agent Skill
Features
Ownership Forensics
Evidence-backed findings with file:line and commit SHAs - force-unwraps, Massive Views, concurrency silencers, secrets, ATS-off, and more.
Control-Loss Score
Scores cognitive debt (N/6), harness maturity (M/5), and salvageability:
Refactor, Stabilize, or Rewrite-core.
Written Report
Outputs IOS_UNVIBE_REPORT.md at the project root with blockers, playbook steps, and a
short chat summary.
Diagnose Only
No silent refactors. Use when you’ve lost the plot on a vibe-coded SwiftUI, UIKit, or AppKit app and need a reclaim plan first.
Install
Add the skill folder
Copy ios-unvibe/ into your agent’s skills directory (path depends on the tool - for
example Cursor uses .cursor/skills or ~/.cursor/skills).
# Example (Cursor)
cp -R ios-unvibe .cursor/skills/ios-unvibe
# or
cp -R ios-unvibe ~/.cursor/skills/ios-unvibeInvoke
Say ios-unvibe in chat. The skill sets disable-model-invocation: true, so
it loads when named.