The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and modify existing Python codebases (SKILL.md). This creates an attack surface where an agent following these instructions might ingest adversarial content (e.g., malicious instructions hidden in code comments or string literals) from a project it is auditing or refactoring.
Ingestion points: Processes local Python source files, configuration files (pyproject.toml), and test files.
Boundary markers: The skill lacks explicit instructions to treat code comments or strings as untrusted data that should not be followed as instructions, though it does discourage the use of docstrings.
Capability inventory: The agent is instructed to execute tooling like uv, ruff, pytest, and prek, and to modify module structure and code logic.
Sanitization: No sanitization or validation of the content of code comments or strings is prescribed before the agent acts upon the code.
[EXTERNAL_DOWNLOADS]: The skill recommends and configures several external dependencies and pre-commit hooks.
Pre-commit hooks: Configures hooks from various GitHub repositories including astral-sh, pre-commit, DavidAnson, tox-dev, pappasam, codespell-project, editorconfig-checker, gitleaks, and compilerla (references/testing-and-tooling.md).
Package Management: Instructs the use of uv to manage dependencies and typer for CLI development.
Trusted Sources: These resources originate from well-known technology organizations and established open-source maintainers within the Python ecosystem.

ac-python