review-pr-branch
Pass
Audited by Gen Agent Trust Hub on Nov 2, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Azure DevOps PR descriptions, work items, and comment threads. These sources can contain malicious instructions designed to influence the agent's review or behavior.
- Ingestion points:
SKILL.md(Workflow Step 2) specifies fetching PR metadata, threads, and linked work-item summaries usingscripts/generate-handoff.pyandazure-devopstools. - Boundary markers: The 'Trust and safety' section explicitly states that PR content should be treated as untrusted data and not instructions. However, no specific technical delimiters (like XML tags or markdown blocks with escaping) are mandated for the prompts themselves.
- Capability inventory: The skill executes local Python scripts (
scripts/parse-git-remote.py,scripts/ensure-env.py,scripts/generate-handoff.py) and uses git commands (git diff,git fetch). It also has network access to Azure DevOps APIs via dependency tools. - Sanitization: The skill mentions that
azure-devopsowns normalization, but does not specify sanitization or escaping of the actual text content before it is processed by the AI for review reasoning. - [COMMAND_EXECUTION]: The workflow involves executing several local Python scripts and Git commands. While these are part of the intended functionality, they represent a surface where input (like branch names or remote URLs) could potentially be manipulated if not handled safely by the underlying scripts.
- Evidence: execution of
scripts/parse-git-remote.py,scripts/ensure-env.py,scripts/generate-handoff.py, andgit diff/git fetchinSKILL.md.
Audit Metadata