The Agent Skills Directory

[CREDENTIALS_SAFE]: The skill manages authentication using environment variables such as AZURE_DEVOPS_PAT, AZURE_DEVOPS_ORG, and AZURE_DEVOPS_TOKEN. It includes explicit instructions to the agent to avoid hardcoding secrets, asking the user to paste them into the chat, or printing them in logs.
[COMMAND_EXECUTION]: The skill uses curl to interact with the Azure DevOps REST API. This behavior is expected and limited to the stated purpose of managing approvals, with specific instructions for handling URL encoding and shell expansion to maintain command integrity.
[EXTERNAL_DOWNLOADS]: The skill communicates with official Azure DevOps domains including dev.azure.com and vssps.dev.azure.com. These are recognized, well-known services for the tasks described.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external API responses, such as pipeline approval instructions and user comments. While this is a potential ingestion surface, the risk is minimized by the structured usage of the data and the primary intended functionality.
Ingestion points: Fields like instructions and comment retrieved from the approvals and builds endpoints in SKILL.md.
Boundary markers: Not explicitly defined for display fields, though output is generally presented in structured tables.
Capability inventory: curl commands restricted to retrieving and updating pipeline states via the Azure DevOps API.
Sanitization: The skill mandates URL encoding for project names and escaping for shell parameters to prevent manipulation of the command structure.

pipelines-approvals