The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several bundled shell scripts (list.sh, auto_assign.sh, report.sh) and a Python script (resolve_job_file.py) to facilitate DevOps workflows. It also invokes uv run issues.py, which executes a script expected to be located in the user's local repository checkout.
[EXTERNAL_DOWNLOADS]: The resolve_job_file.py script utilizes the PEP 723 metadata format to declare a dependency on PyYAML. This package is downloaded from the official Python Package Index (PyPI) at runtime, which is a well-known and standard package registry.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests untrusted data from Azure DevOps and has the capability to modify the local environment based on that data.
Ingestion points: Ticket titles and descriptions containing potentially untrusted content are retrieved from Azure DevOps via uv run issues.py get <id>.
Boundary markers: There are no explicit delimiters or markers used to isolate the retrieved ticket content from the agent's internal instructions.
Capability inventory: The skill can perform network write operations to Azure DevOps (for ticket assignment) and modify local source files in the repository (for applying suggested fixes).
Sanitization: No validation or sanitization of the retrieved ticket content is performed before it is used to inform code changes or summaries.
Mitigation: The Python script resolve_job_file.py correctly uses yaml.safe_load() to process configuration files, preventing unsafe deserialization attacks.

fabricks-devops-issues