git-remote-pr

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the git and GitHub CLI (gh) tools to perform repository state collection and pull request management operations.
  • Evidence: Found in scripts/prepare-pr.ps1 and scripts/execute-pr.ps1 scripts.
  • [DYNAMIC_EXECUTION]: The integration test harness dynamically generates and executes PowerShell code via the -EncodedCommand parameter to facilitate failure injection and mocking.
  • Evidence: scripts/test-pr.ps1 contains a Base64-encoded script block passed to a new PowerShell process.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content, such as commit messages and code diffs, to generate pull request descriptions.
  • Ingestion points: scripts/prepare-pr.ps1 collects data from git log and git diff outputs.
  • Boundary markers: The skill implements strict structural enforcement in scripts/pr-common.ps1 (Assert-PrBodyStructure) to ensure the agent's generated description conforms to a safe schema.
  • Capability inventory: The skill's write capabilities are scoped to git push and gh pr metadata updates.
  • Sanitization: It actively filters non-portable content like diffhunk:// links and maps file changes to reviewer-oriented themes.
  • [PRIVILEGE_ESCALATION]: The test suite adjusts file execution permissions to create a mock environment for the GitHub CLI.
  • Evidence: scripts/test-pr.ps1 executes chmod +x on a dynamically created shim script in a temporary directory.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 04:02 PM