The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data retrieved from application error logs and user feedback, which could potentially contain malicious instructions aimed at the agent.
Ingestion points: [Amplitude] Error Logged (specifically the Error Message and Error Stack Trace properties) and user feedback insights retrieved via the Amplitude:use_amplitude_ai_feedback tool.
Boundary markers: The instructions do not define explicit delimiters or instructions to the agent to treat the content of error logs or feedback as data only, increasing the risk of the agent following embedded instructions.
Capability inventory: The skill uses Amplitude-specific tools for querying data (Amplitude:query_amplitude_data), retrieving session replays (Amplitude:get_amp_session_replay_info), and checking feature flags/experiments. It does not possess capabilities for arbitrary shell execution, file system modification, or non-Amplitude network operations.
Sanitization: There are no instructions for the agent to sanitize or validate the content of error messages or feedback strings before processing them.

diagnose-errors