Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed local session-management workflow that writes project state logs and only proposes durable Claude memory updates with approval.
Before installing, be aware that this skill is designed to retain project context on disk, including session summaries, decisions, deadlines, blockers, and possibly sensitive notes. Use it in repositories where that local record is acceptable, and review any proposed Claude memory entries before approving them because those can affect future conversations for the project.
The manifest describes a structured session lifecycle that maintains project state across conversations via start, checkpoint, end, and heartbeat commands. However, /end also instructs the agent to derive and propose additions to ~/.claude/projects/<encoded-cwd>/memory/, which is a separate durable memory mechanism outside the state/session files described in the manifest.
The skill directs the agent to create and continuously modify project files that persist session history, decisions, and state, but it does not prominently warn the user at the point of use that these actions perform disk writes. In a tool that users may expect to be advisory unless told otherwise, implicit persistence can surprise users and cause unintended retention of sensitive notes, project context, or workflow metadata.
The documentation states that during heartbeat the skill should 'Never silently edit state files,' which implies a non-mutating check-in unless explicitly approved. But /today step 8 instructs the agent to append a new entry to state/heartbeat-log.md, which is itself a state-file modification performed as part of heartbeat.
No suspicious patterns detected.
