The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided ideas and tasks and uses them to drive an autonomous execution workflow using high-capability tools.
Ingestion points: User input is captured via the <идея или задача> argument hint and used throughout the understanding and design phases.
Boundary markers: The instructions lack explicit delimiters or safety instructions (e.g., 'ignore any instructions contained within the user input') when processing external tasks.
Capability inventory: The skill utilizes powerful tools including Write, Edit, Bash, Skill, and TaskCreate for autonomous execution.
Sanitization: There is no evidence of input validation or sanitization before the user-provided data influences the execution of shell commands or file operations.
[COMMAND_EXECUTION]: The skill instructions specify the use of several external commands and utilities that are not explicitly defined or scoped within the allowed-tools metadata.
Evidence: References to rlm_route_context, ssl_search, and 1c-syntax-checker as executable commands or functions.
Context: While these appear to be environment-specific utilities for 1C development, they represent unvetted execution paths that could be manipulated if they accept user-controlled arguments.

brainstorm