The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest untrusted user requirements and codebase paths which are passed to subagents. These subagents have the authority to modify the filesystem and execute commands, creating a potential vector for indirect prompt injection attacks.
Ingestion points: The skill explicitly instructs agents to ingest "user requirements verbatim" and "existing codebase paths" in the TDD, Feature Addition, and Debugging workflows.
Boundary markers: There are no specific instructions for applying boundary markers or delimiters when interpolating untrusted requirements into agent prompts.
Capability inventory: The workflow delegates to agents like python-cli-architect which can write code to files, and includes validation steps that execute shell commands such as uv run pytest and bandit.
Sanitization: The instructions do not specify any sanitization or validation logic for the user-supplied requirements before they are processed by the planning or implementation agents.
[COMMAND_EXECUTION]: The skill guides the orchestrator to have agents execute various shell commands for development automation and quality assurance.
Evidence: The workflows include commands for package management and testing using uv, specifically uv run pytest for test execution, uv run bandit for security scanning, and uv run mutmut for mutation testing.
Evidence: The skill uses the /python-engineering:shebangpython command, which is described as directly editing files to ensure PEP 723 compliance.

orchestrating-python-development