The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from user inputs (such as bug reports, feature descriptions, and maintenance requests) and incorporates this content into the shared state or passes it to downstream skills.
Ingestion points: User-provided text in trigger phrases (e.g., "There's a bug in production: ") and project-level files like CONTEXT.md and .lcs/state.md.
Boundary markers: The skill documentation does not specify the use of delimiters or sanitization routines for user-supplied descriptions before routing.
Capability inventory: The skill can write to the local filesystem (creating .lcs/state.md and session-log.md) and possesses the capability to invoke other skills in the LCS suite.
Sanitization: No explicit evidence of input sanitization or validation of the user-provided content was found.
[COMMAND_EXECUTION]: The skill is designed to dynamically select and invoke downstream skills based on user intent. While this is the intended functionality of an orchestrator, it involves the execution of external logic (other agent skills) driven by natural language input.

lcs-master