The reviewed behaviors match Atris's stated CLI and agent-security purpose. The suspicious scanner leads resolve to an opt-in guard, constrained event reporting, and code that detects invisible characters.
Static reason
High-risk behavior combination matched malicious policy.; previous stored version diff introduced dangerous source
Trigger
CLI features run when invoked; the Sentinel hook runs after a user explicitly installs it and Claude Code calls the hook.
Impact
No harmful behavior was identified. Guard events omit command text and file paths.
Mechanism
The optional guard checks agent actions, records limited block-event metadata, and can send queued events to the user's Atris dashboard when an Atris token is available.
Evidence
<project>/.claude/hooks/sentinel-guard.js<project>/.claude/settings.json~/.claude/hooks/sentinel-guard.js~/.claude/settings.json~/.atris/sentinel/guard.log~/.atris/sentinel/queue.jsonl
Network endpoints1
api.atris.ai/api/sentinel/events