This version is not clearly malicious: AI tracking is part of the assessment CLI, is disclosed before setup, and hosted capture is gated by an assessment backend lease. However, init changes user-level AI hooks and trust settings, so the broader agent configuration change merits a warning.
Static reason
One or more suspicious static signals were detected.
Trigger
User runs `litmus init` for an assessment.
Impact
AI prompts, responses, and tool activity from supported tools may be captured outside the assessment folder. This broad agent-configuration change is disclosed and serves the package's assessment purpose, but warrants a warning.
Mechanism
The command writes user-scope AI hooks and, for Codex, pins hook trust; hosted capture uses an assessment token and server-issued lease, then batches captured activity to the assessment backend.
Evidence
dist/commands/init.jsdist/lib/hosted-capture.jsdist/lib/hook-logger.cjs~/.claude/settings.json~/.codex/hooks.json~/.codex/config.toml~/.litmus/hosted-capture.json<assessment>/.litmus/activity.jsonl