azure-devops
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from Azure DevOps (such as work item descriptions and pull request comments) which may contain malicious instructions for the agent or downstream consumers.
- Ingestion points:
fetch-work-item.pyandfetch-pr-comments.pyretrieve text content from the Azure DevOps REST API. - Boundary markers: The skill normalizes all data into a strictly defined JSON handoff contract (
assets/schemas/ado-openspec-handoff.schema.json) to isolate the untrusted content. - Capability inventory: The skill possesses write capabilities (creating work items, comments, and pull requests) via scripts like
create-work-item.py, though these are explicitly gated by mandatory dry-run outputs and user confirmation flags. - Sanitization:
normalize-work-item.pyuses an HTML parser to strip tags from description and comment fields, and_ado_lib.pyimplements recursive redaction to prevent tokens from leaking in output. - [COMMAND_EXECUTION]: Orchestration scripts execute internal Python utilities to perform multi-stage data processing.
- Evidence:
generate-handoff.pyandemit-handoff.pyusesubprocess.runto call sibling scripts such asfetch-work-item.pyandvalidate-handoff.py. These calls usesys.executableand structured argument lists (withshell=False) to ensure local execution context and prevent injection. - [DATA_EXFILTRATION]: The skill handles sensitive Personal Access Tokens (PATs) and implements controls to prevent unauthorized data transfer.
- Evidence:
_ado_lib.pycontains a strict URL allowlist (dev.azure.com,vssps.dev.azure.com, and*.visualstudio.com) for all authenticated requests. This restricts the agent's ability to send data or credentials to non-Azure DevOps domains.
Audit Metadata