create-rc-tag

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes standard Git commands to manage repository tags and branches.
  • Evidence includes: git ls-remote, git fetch, git tag -s, and git push operations.
  • Safety measures: The skill implements strict regex validation for user-provided inputs (BASE_VERSION and RC_NUMBER) to prevent command injection. It targets the official NVIDIA/NeMo-Relay repository, uses signed tags (-s) for authenticity, and explicitly requires user approval before pushing changes.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied version strings, which represents a potential injection surface.
  • Ingestion points: BASE_VERSION and RC_NUMBER variables in SKILL.md.
  • Boundary markers: Strict Bash regex markers (^...$) are used to delimit inputs.
  • Capability inventory: The skill has the capability to write to a remote repository via git push.
  • Sanitization: Inputs are rigorously validated against numeric patterns (^[0-9]+\.[0-9]+...$), effectively neutralizing the risk of malicious instruction injection through these fields.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 05:19 PM
Security Audit — agent-trust-hub — create-rc-tag