session-hooks

Warn

Audited by Gen Agent Trust Hub on Jul 1, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill configures project-level hooks in .claude/settings.json that execute shell commands (infynon trace, jq, cat) on session start and stop events.
  • [REMOTE_CODE_EXECUTION]: The documentation instructs users to execute a bash script from a variable path (install.sh), which facilitates the execution of unverified external code.
  • [DATA_EXFILTRATION]: Automated hooks trigger data synchronization to a remote service (infynon trace sync --direction push), potentially uploading sensitive session observations without explicit per-action user review.
  • [PROMPT_INJECTION]: The shell hooks produce output containing control strings (e.g., [TRACE-HOOK]) and direct instructions (e.g., Invoke @tracer agent) that guide agent behavior, serving as an indirect prompt injection surface.
  • [EXTERNAL_DOWNLOADS]: The skill automates the pulling of data from remote sources into the local context via the sync command during session initialization.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 1, 2026, 07:15 AM
Security Audit — agent-trust-hub — session-hooks