The Agent Skills Directory

[DATA_EXPOSURE]: The skill uses the AZURE_DEVOPS_PAT environment variable for authentication with a well-known service (dev.azure.com). It contains proactive security instructions directing the agent to only read the token from environment variables and never to print it, write it to a file, or ask the user to provide it in plain text during a session.
[COMMAND_EXECUTION]: The skill uses curl to interact with APIs. It includes defensive measures to prevent shell injection, such as mandatory URL encoding for project and pipeline names, and escaping special characters like $ to avoid shell expansion. Additionally, for Windows PowerShell, it specifies writing JSON bodies to temporary files to avoid command-line parsing issues.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from Azure DevOps API responses and user-supplied parameters.
Ingestion points: User-provided inputs (project names, pipeline names, branch names, runtime parameters) and external API responses (build definitions, pipeline run lists, and execution timelines).
Boundary markers: No explicit natural language boundary markers are used in the prompt construction.
Capability inventory: Shell command execution via curl, network access to Azure DevOps, and file-write access for temporary data storage (PowerShell body.json).
Sanitization: The skill mandates URL encoding for dynamic path/query components and uses structured JSON for data transmission, which mitigates the risk of instructions embedded in data being misinterpreted as commands.

pipelines-run