setup-pipeline

Pass

Audited by Gen Agent Trust Hub on Oct 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • Command Execution: The skill utilizes the Bash tool to execute local utility scripts and Microsoft CLI tools such as az (Azure CLI), pac (Power Platform CLI), and git. These operations are used to manage project state, handle authentication, and commit configuration changes to the repository.
  • Credential Management: It retrieves OIDC and Bearer access tokens for Power Platform services using az account get-access-token. These tokens are generated at runtime and passed to utility scripts to authorize API calls against Dataverse environments.
  • Environment Configuration (Privilege Operations): The skill identifies and offers to modify the blockedattachments security setting in Power Platform environments. This change is necessary to allow the deployment of JavaScript files in code-based sites. This operation is governed by a specific user consent gate (Phase 4.4) to ensure the user is aware of the environment-level impact.
  • External Data Grounding: The skill employs MCP tools to search and retrieve content from Microsoft's official documentation site (Microsoft Learn). This information is used to verify API schemas and ensure alignment with current platform requirements.
  • Indirect Prompt Injection Surface: The skill processes several data sources that could potentially contain untrusted instructions if compromised.
  • Ingestion points: Local project files (powerpages.config.json, .solution-manifest.json) and live data returned from Dataverse OData APIs.
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when the agent processes content from these files.
  • Capability inventory: The skill possesses extensive capabilities, including shell command execution via Bash, file system modification via Write/Edit, and network access through Node.js utility scripts.
  • Sanitization: The instructions do not detail specific sanitization or validation logic for the content ingested from local configuration files or API responses, though the operations are generally structured around technical metadata (GUIDs and URLs).
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 25, 2026, 04:02 AM