The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted source code provided via the $ARGUMENTS path. This creates a vulnerability where instructions embedded in the code (e.g., in comments or string literals) could be executed by the agent.
Ingestion points: Source code files read from the file system using Read, Grep, or Glob based on the user-provided path.
Boundary markers: None. The instructions do not specify that the agent should ignore or treat the audited file content as untrusted data.
Capability inventory: The skill allows the agent to read files, search for patterns, and execute restricted shell commands (nx, wc).
Sanitization: There is no evidence of sanitization or filtering of the file content before it is processed by the agent.

best-practices