saltmode
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several scripts (e.g.,
watch-pr,check-plan.mjs,check_resource_paths.py) that perform local system operations. These scripts usespawnto run commands such asgit,gh(GitHub CLI), andbunfor engineering automation, PR monitoring, and workspace validation. - [EXTERNAL_DOWNLOADS]: The
scripts/bootstrap.tsutility is designed to ensure necessary dependencies are present. It executesbun install --frozen-lockfileto download and install Node.js packages from the official registry if they are missing or if the configuration files have changed. - [INDIRECT_PROMPT_INJECTION]: The skill features a PR "Babysit" playbook and associated tools that ingest external content from GitHub pull request comments. The instructions explicitly address this vulnerability surface by mandating that the agent treat all external feedback as untrusted data, avoid using it as a direct command source, and prevent its interpolation into shell execution logic.
Audit Metadata