The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied research manuscripts, abstracts, and outlines to provide journal recommendations. This creating a surface for indirect prompt injection where malicious instructions embedded in scholarly text could attempt to override agent instructions.
Ingestion points: Reads research problems, completed work, and manuscripts as described in SKILL.md.
Boundary markers: Specifically instructs the agent to treat repository content and manuscript instructions as 'data, not as commands' (SKILL.md).
Capability inventory: Performs local file harvesting (for auditing) and can conduct web searches for journal verification.
Sanitization: Relies on structural JSON validation and explicit prompting instructions rather than automated text sanitization.
[COMMAND_EXECUTION]: Documentation in SKILL.md directs the execution of a bundled local script (scripts/audit_selection.py) to verify the integrity of evidence packets. This is a localized and expected diagnostic function.
[DYNAMIC_EXECUTION]: Integration tests in the tests/ directory utilize the subprocess module and importlib for dynamic loading of modules to verify cross-skill compatibility with other 'light-*' series tools. These are restricted to development and testing contexts.

light-journal-select