The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and ingest local codebase content, which represents a potential surface for indirect prompt injection from untrusted source files.
Ingestion points: The skill instructions specify using Glob to survey source roots and Read or Grep to inspect module implementations as part of its diagnostic process.
Boundary markers: The instructions do not provide specific delimiters or ignore-instructions warnings when reading external content from the file system.
Capability inventory: The skill is granted file system access through Read, Grep, Glob, and Write tools as listed in the allowed-tools configuration.
Sanitization: No explicit data sanitization or validation protocols are mentioned for the content retrieved from the project files.

architecture-rescue