The Agent Skills Directory

[PROMPT_INJECTION]: The script scripts/query_manual.py is vulnerable to indirect prompt injection. User-supplied input from the question argument is directly interpolated into a system prompt for the Gemini API using f-strings. This allows a user to potentially bypass the established 'Strict Rules' and influence the behavior of the chip manual query assistant.
Ingestion points: The question command-line argument in scripts/query_manual.py.
Boundary markers: While the manual content is delimited, the user-provided question is included following a simple label without any escaping or isolation from the instructions.
Capability inventory: The skill environment has access to powerful tools including Read, Write, Edit, and Bash. The query script itself performs network requests to the Gemini API.
Sanitization: No input validation, escaping, or sanitization is performed on the user's question before it is used to construct the prompt.

chip-manual