The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data in the form of client feedback, complaints, and feature requests. While it includes safety instructions, the combination of external data processing and codebase modification capabilities creates an inherent attack surface.
Ingestion points: Untrusted data enters the agent context through the input_gate which accepts raw feedback text, tickets, or chat messages from external clients.
Boundary markers: The skill contains explicit instructions in the input_gate section to treat pasted messages as untrusted data and explicitly warns not to execute commands, follow links, or expose secrets requested within the client text.
Capability inventory: The skill has authority to perform repository inspection, search, codebase editing, and execution of local validation commands as part of its verification workflow.
Sanitization: The instructions require the agent to separate observed facts from assumptions and cite original feedback, but do not specify technical sanitization methods beyond logical separation and the aforementioned input gate warning.

client-feedback