The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided code, configuration files, and fetch external data, creating an ingestion surface for potentially malicious instructions.
Ingestion points: Processes user code implementation requests, reads local configuration files in the read_config example, and fetches data from external URLs in the fetch_all example.
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the data it processes.
Capability inventory: The skill utilizes httpx for network requests, pathlib for file system interactions, and standard linting/type-checking tools (mypy, ruff, black) which involve subprocess execution.
Sanitization: No explicit sanitization or input validation logic is defined for the external data processed at runtime.
[EXTERNAL_DOWNLOADS]: The skill references numerous standard Python libraries and development tools across its documentation and examples.
Evidence: Recommends well-known packages including httpx, aiofiles, pytest, mypy, black, ruff, pydantic, hypothesis, and syrupy. These references target official registries and are standard for the described use case.
[METADATA_POISONING]: There is a discrepancy in the author identity within the skill metadata.
Evidence: The SKILL.md frontmatter identifies the author via a GitHub profile for 'Jeffallan', which differs from the provided vendor context 'alismx'. This inconsistency is documented as a metadata finding but does not appear to be malicious.

python-pro