rhdh-pr-create

Pass

Audited by Gen Agent Trust Hub on Aug 19, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Jira, GitHub issues, and previous skill outputs. It mitigates injection risks through the following:\n
  • Ingestion points: Change handoff file lists, issue references, and recordings from /rhdh-plugin-bug-fix.\n
  • Boundary markers: Mandatory 'Approval Gates' for every external write and the requirement to 'State every write' in full with a preview.\n
  • Capability inventory: Interaction with local git, gh CLI, and yarn.\n
  • Sanitization: Uses HEREDOCs for multi-line string assembly and verifies that uploaded recording URLs start with a trusted prefix (https://raw.githubusercontent.com).\n- [DATA_EXPOSURE_AND_EXFILTRATION]: Protects against leaking sensitive local configuration by explicitly excluding pre-existing dirty files and local environment variables from the staged publication set.\n- [DYNAMIC_EXECUTION]: Employs a short, static Python snippet for JSON parsing and asset URL validation. The script is used strictly for utility purposes and does not execute untrusted logic.\n- [PRIVILEGE_ESCALATION]: Prohibits the use of sudo within the skill's instructions, ensuring all operations occur with the agent's standard user permissions and delegating permission fixes to the human user.\n- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill runs yarn to install dependencies and build packages. This is a standard functional requirement for publishing changes to the targeted repositories.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 19, 2026, 11:11 AM
Security Audit — agent-trust-hub — rhdh-pr-create