The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted external data from meeting transcripts and summaries which could contain adversarial instructions.
Ingestion points: Untrusted content is ingested from transcript.json and 회의요약.txt (or geminiまとめ.txt) as specified in the skill's input section and argument hints.
Boundary markers: Absent. The instructions do not define delimiters or specific warnings for the agent to ignore any instructions found within the transcript or summary text.
Capability inventory: The skill identifies and reads input files and specifies a local file-write operation to meeting_rec/transcribe/topics.json. It does not demonstrate network access or arbitrary code execution capabilities.
Sanitization: While the skill includes internal validation steps for the output (e.g., verifying timestamps and topic counts), it does not perform sanitization or filtering of the text content to prevent embedded instructions from influencing the agent.

topics-extractor