The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted code from files and directories to provide explanations.
Ingestion points: Uses Read(*), Glob(*), and Grep(*) to read project files as specified in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore instructions embedded within the code being analyzed (e.g., malicious comments).
Capability inventory: The skill has access to WebSearch, WebFetch, and restricted Bash (git) commands.
Sanitization: No sanitization or validation of the ingested code is performed before the agent processes it for explanation.
[DATA_EXFILTRATION]: The skill instructions in SKILL.md direct the agent to identify library versions from dependency files (like package.json or pom.xml) and then use WebSearch and WebFetch to look up documentation. While library metadata is typically non-sensitive, this pattern creates a functional path where data read from the local environment is sent to external web services.

explain-code