orbit
Pass
Audited by Gen Agent Trust Hub on Oct 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill's primary function is to generate and execute shell scripts (
run-loop.sh,verify.sh,bootstrap.sh,recover.sh,notify.sh) to manage autonomous agent loops. - Evidence: Script templates are defined in
reference/script-template-runner.mdandreference/script-template-support.md. - [DYNAMIC_EXECUTION]: The generated runner uses
bash -cto execute a configurableEXEC_CMDand a portable timeout utility implemented viaperl -ecode. - Evidence: The
run_with_budgetandportable_timeoutfunctions inreference/script-template-runner.md. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted agent output from tool calls and files like
done.mdto generate git commit messages and update loop progress, creating a potential injection surface. - Evidence: The
squash_iteration_branchfunction inreference/script-template-runner.mdcaptures LLM output for use ingit commitcommands. - Mitigation: The skill implements robust defenses including SHA-256 pinning of goal files (
GOAL_IMMUTABLE), verification scripts (ARTIFACT_PINS), and agent settings (.claude/settings.json), alongside a triple-gate verification process to ensure tasks are completed correctly and without malicious stubs. - [EXTERNAL_DOWNLOADS]: The
notify.shtemplate references theedge-ttsutility for text-to-speech notifications. - Evidence: Found in the
notify.shtemplate withinreference/script-template-support.md.
Audit Metadata