orbit
Warn
Audited by Gen Agent Trust Hub on Apr 10, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill generates several bash scripts like run-loop.sh and bootstrap.sh that perform system-level operations including file manipulation and process management.
- [COMMAND_EXECUTION]: The core logic involves dynamically assembling and executing shell commands. The run-loop.sh template uses a portable_timeout function implemented via perl to execute an AI engine command EXEC_CMD defined in the configuration.
- [PROMPT_INJECTION]: The skill ingests untrusted data from sources like goal.md and git commit summaries. Ingestion points: bootstrap.sh and run-loop.sh read goal.md; notify.sh reads git diffs. Boundary markers: None. Capability inventory: Shell command execution, file writes, and git operations. Sanitization: No evidence of escaping or validation of external content.
- [EXTERNAL_DOWNLOADS]: The notify.sh script references edge-tts, which communicates with external Microsoft services.
- [COMMAND_EXECUTION]: The references/executor-engines.md file encourages using high-privilege flags for external AI tools, such as --yolo for Gemini and --dangerously-skip-permissions for Claude, which explicitly bypass safety confirmations and grant significant autonomy over the local environment.
Audit Metadata