The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core rules (SKILL.md, Rule 26) instruct the agent to execute development tools like pytest and ruff on the user's project codebase. This creates a surface for indirect prompt injection where malicious code or instructions embedded in the user's files (e.g., in conftest.py or test logic) could be executed by the agent during the suggested validation process.
Ingestion points: User-provided Python source files, test files, and project configuration files.
Boundary markers: No explicit boundary markers or warnings to ignore embedded instructions are defined for the agent's code analysis.
Capability inventory: Execution of shell commands (uv run pytest, uv run ruff) that dynamically load and run project-defined code and plugins.
Sanitization: The instructions do not include steps for environment isolation or content sanitization before tool execution.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation of the httpx2 package as a test dependency (references/testing.md, references/tooling.md). httpx2 is not a standard or officially recognized successor to the httpx library in the current Python ecosystem. The recommendation is based on claims regarding future versions of FastAPI and Starlette that are not currently available. Recommending unknown or non-standard packages from public registries can lead to supply chain attacks such as typosquatting or dependency confusion.
[DYNAMIC_EXECUTION]: The workflow for testing and linting (SKILL.md, Rule 26) involves the dynamic execution of project code through the pytest framework. While this is a standard developer activity, it represents a mechanism for untrusted code execution within the agent's operating context.

fastapi