The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module to execute claude and copilot CLI tools. These executions are core to the skill's purpose: performing automated evaluations of skill descriptions and running optimization loops to improve trigger rates.
Evidence: Found in scripts/run_eval.py, scripts/improve_description.py, and scripts/run_loop.py.
[COMMAND_EXECUTION]: The scripts/generate_review.py utility executes lsof to manage port availability for its local visualization server.
[DATA_EXPOSURE]: The skill includes a local visualization tool (scripts/generate_review.py) that starts a tiny HTTP server on 127.0.0.1 to serve evaluation reports and artifacts to the developer's browser. This is a standard developer workflow pattern for results visualization.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it processes external SKILL.md files provided by the user to generate Azure deployment wrappers.
Ingestion points: scripts/scaffold_azure_agent.py reads user-specified SKILL.md files.
Boundary markers: Absent; the content is interpolated directly into templates.
Capability inventory: The skill can write files to the local disk and execute claude CLI commands via subprocesses.
Sanitization: Uses Python string templates and .format() for code generation.
[DYNAMIC_EXECUTION]: The skill dynamically generates Python orchestrators and Bicep infrastructure templates using Jinja2 or string templates. It also dynamically creates and deletes Markdown files in the .claude/commands/ directory during evaluation phases to test model trigger behavior.

create-azure-agent