custom-skills-upstream-ops
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/check_uds.pyscript executes localgitcommands usingsubprocess.run. The implementation uses list-based arguments without a shell, which effectively prevents command injection vulnerabilities. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external commit logs and file contents, creating an attack surface for indirect prompt injection. 1. Ingestion points: Commit messages and file content are read in
modes/audit.mdandmodes/overlap.md. 2. Boundary markers: Absent. 3. Capability inventory: File writes to local report directories inscripts/check_uds.pyandmodes/maintenance.md, and shell command execution inscripts/check_uds.py. 4. Sanitization: Absent. The risk is considered negligible due to mandatory manual user review for all suggested actions. - [SAFE]: All operations are scoped to local configuration and repository management within the project's environment. The skill follows best practices by requiring user confirmation for synchronization updates and report archiving.
Audit Metadata