The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources within the Amplitude platform, which could potentially contain malicious instructions designed to manipulate the agent's analysis or output.
Ingestion points: The skill instructions involve calling get_amplitude_context to read the AI context field (which can contain arbitrary business context) and use_amplitude_ai_feedback to fetch customer insights, mentions, and comments.
Boundary markers: The instructions do not include the use of delimiters (e.g., XML tags, triple backticks) or explicit framing to prevent the agent from following instructions that might be embedded within the retrieved feedback or context data.
Capability inventory: The skill is configured to use a suite of data-retrieval tools (charts, funnels, experiments, feedback) to synthesize reports. While it does not demonstrate capabilities for arbitrary shell execution or file system modification, the synthesis logic could be biased or compromised by the ingested data.
Sanitization: No mention is made of sanitizing or filtering the content of the customer feedback or AI context fields before they are processed by the agent.

discover-opportunities