The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by ingesting and processing external data during the instrument design process.\n
Ingestion points: The skill is designed to take 'Measurement targets' and 'Existing instruments' as inputs, including reading files from the local filesystem (e.g., via the edit-preservation protocol).\n
Boundary markers: There are no explicit delimiters or safety instructions provided to the agent to treat these external inputs as untrusted data, which could allow embedded instructions in a measurement target or survey file to influence the agent's behavior.\n
Capability inventory: The skill has the ability to write files to the system using the ane_package.reporting.word_export and xlsform.write_xlsform tools.\n
Sanitization: No specific validation or sanitization logic is described to filter malicious content from the measurement targets or existing instruments before they are processed.

instrument-designer