Security audit
Security checks for vulnerabilities and agentic risk
This is a disclosed, purpose-aligned checkpoint skill that advises on risk and escalation, with local audit logging and fail-open behavior users should understand.
Before installing, decide whether you are comfortable with checkpoint audit files being retained under ~/.claude and with workflows continuing if this checkpoint cannot run. For regulated or highly sensitive projects, configure the calling commands to fail closed or prompt the user on checkpoint failure, and periodically review or delete stored audit logs.
The failure-handling section instructs calling commands to continue execution if checkpoint invocation fails, effectively converting a safety gate into fail-open behavior. In high-stakes review or escalation workflows, an attacker or fault condition could disable the checkpoint path and cause risky actions to proceed without reversibility assessment or required user confirmation.
The skill specifies persistent local logging of checkpoint context, affected files, confidence, and session details under user home directories, but does not require notice, consent, minimization, or redaction. In an agent setting, this can silently retain sensitive project metadata, filenames, architectural decisions, and workflow context that may expose confidential information to other local users, backups, or later tooling.
No suspicious patterns detected.
