document-release
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from Pull Request (PR) or Merge Request (MR) bodies, which creates a surface for indirect prompt injection.
- Ingestion points: PR/MR bodies are fetched using platform-specific commands (
gh pr view,glab mr view) in Step 9. - Boundary markers: The skill uses a "trust envelope" and the
gstack-issue-guardtool to wrap external content. - Capability inventory: The skill has access to file system tools (
Read,Write,Edit,Glob), shell execution (Bash), and source control tools (git,gh,glab). - Sanitization: Content is passed through
gstack-redactfor privacy and a "banner tripwire" mechanism in Step 9 to prevent the agent from following instructions embedded in the external content. - [EXTERNAL_DOWNLOADS]: The skill may prompt the installation of the
@openai/codexpackage from the official npm registry if it is not present. This is a well-known tool from a recognized service provider. - [COMMAND_EXECUTION]: The skill executes various Git and platform CLI commands to manage documentation and PR state, consistent with its primary function.
- [DYNAMIC_EXECUTION]: The skill uses
gstack-skill-startto receive and execute one-time instruction blocks for onboarding and consent, guarded by matching session IDs.
Audit Metadata