autonomous-dispatcher
Warn
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONOBFUSCATION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses a wide array of system utilities including gh, git, aws ssm, jq, and openssl to interact with code hosts and manage local and remote agent environments.
- [PRIVILEGE_ESCALATION]: Scripts like scripts/lib-ssm.sh and scripts/dispatch-remote-aws-ssm.sh construct commands using sudo -u to execute tasks as a specific target user on remote machines via AWS Systems Manager.
- [PERSISTENCE]: The scripts/install-gc-timer.sh script installs persistent background timers using crontab on Linux and launchd on macOS to run the adt-gc.sh garbage collection utility.
- [DYNAMIC_EXECUTION]: The skill employs eval in scripts/dispatcher-multi-tick.sh to process project-specific configuration blocks and in scripts/lib-turn-limit.sh to tokenize launcher arguments.
- [INDIRECT_PROMPT_INJECTION]: The system processes untrusted input from GitHub and GitLab issues. Although it employs boundary markers and explicit instructions to ignore commands within the issue content, the ingestion of untrusted data into an agent with broad command execution capabilities creates an inherent risk surface.
- [OBFUSCATION]: The scripts/lib-ssm.sh script uses Base64 encoding to wrap shell commands sent to remote hosts over AWS SSM, which ensures transport safety but also masks the literal command string from simple text monitoring during transmission.
Audit Metadata