The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by instructing the agent to ingest and analyze external, untrusted content (source code, scripts, and client components) during its review process.
Ingestion points: The SKILL.md file defines a "Code Review" task that requires reading "scripts, client components, and browser execution paths."
Boundary markers: There are no explicit instructions or delimiters provided to separate the untrusted code being reviewed from the agent's internal instructions.
Capability inventory: The process relies on file-read capabilities to inspect the target codebase.
Sanitization: The skill does not define methods for sanitizing or escaping the content of the reviewed files to prevent embedded instructions from influencing agent behavior.

json-safety