The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill implements strong data protection policies, explicitly instructing the agent to unconditionally refuse real secrets, live credentials, customer data, and production data. It further mandates the redaction of sensitive information, such as private paths and authorization headers, from all generated reports and logs.
[REMOTE_CODE_EXECUTION]: The skill identifies and mitigates the risk of arbitrary code execution via type-checker plugins. It mandates a refuse-or-ask policy for any unreviewed plugins in the project configuration. Additionally, the included scripts/scan_annotations.py utility is a pure static analyzer that uses the ast module to parse code without execution, following secure coding practices like input validation and resource limits.
[INDIRECT_PROMPT_INJECTION]: The skill recognizes the risk of processing untrusted data from the target repository. It includes specific instructions to treat repository content, logs, and generated output as untrusted and explicitly forbids following any instructions embedded within that data.
[COMMAND_EXECUTION]: The skill manages the execution of type-checking tools (mypy, pyright) safely by prioritizing existing project configurations and requiring explicit user approval before installing new tools or environment modifications.
[DYNAMIC_CONTEXT_INJECTION]: The skill does not utilize dynamic context injection patterns (e.g., shell command execution at load time).

python-type-safety