The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The fetch_url function in modules/real-world-applications.md fetches content from external URLs using aiohttp and returns the raw text to the agent's context.
Ingestion points: The response.text() call in modules/real-world-applications.md is a direct ingestion point for untrusted external data.
Boundary markers: Absent. The skill does not provide instructions or delimiters to isolate the fetched content or warn the agent to ignore instructions embedded within it.
Capability inventory: As a Python development skill, it is used in environments where the agent likely has access to tools like bash and file system operations.
Sanitization: Absent. There is no evidence of filtering, escaping, or validation of the fetched content before it is processed.
[EXTERNAL_DOWNLOADS]: The modules/testing-async.md file includes instructions to install pytest-asyncio via pip. This is a standard and well-known library within the Python ecosystem, appropriate for the skill's stated purpose of teaching async testing.
[COMMAND_EXECUTION]: The get_user_data function in modules/real-world-applications.md demonstrates an insecure pattern for database queries by using f-strings for SQL construction (e.g., f"SELECT * FROM users WHERE id = {user_id}"). If an agent adopts this pattern for real-world code generation with user-provided input, it would introduce SQL injection vulnerabilities.

python-async