The Agent Skills Directory

[COMMAND_EXECUTION]: The create-backlog.cs script programmatically spawns the az (Azure CLI) process using System.Diagnostics.Process to obtain Entra access tokens for Azure DevOps authentication via the az account get-access-token command.
[DYNAMIC_EXECUTION]: The skill operates by executing C# source code at runtime via dotnet run, which involves dynamic compilation and execution of scripts stored within the skill's distribution.
[PRIVILEGE_ESCALATION]: Documentation in SKILL.md and comments in create-backlog.cs instruct users to execute PowerShell scripts using the -ExecutionPolicy Bypass flag, which is a technique used to circumvent local security restrictions on script execution.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface for indirect prompt injection as it processes backlog_input.json, which is intended to be generated from external findings that may contain untrusted data.
Ingestion points: The create-backlog.cs script ingest data from backlog_input.json (or a path specified by the user/agent) using the JsonDocument.Parse method.
Boundary markers: A mandatory dry-run gate is implemented by default (AZDO_DRY_RUN="true"), and the agent is explicitly instructed to seek human confirmation before initiating the real run.
Capability inventory: The skill can perform authenticated HTTP PATCH requests to Azure DevOps, execute local shell commands, and write result files to the local file system.
Sanitization: The script uses Uri.EscapeDataString for work item types and includes warnings in SKILL.md about the necessity of escaping HTML special characters in description fields.

ado-create-work-items