The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands such as mv and grep to organize transcript files into task-specific directories and to index section headers. These operations are localized to the workspace and serve the skill's primary purpose of data organization.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of meeting transcripts, which presents an inherent surface for indirect prompt injection.
Ingestion points: Reads raw content from user-provided .txt files (e.g., original.txt).
Boundary markers: No explicit XML/fence delimiters are defined to separate transcript content from instructions, though the workflow is step-based.
Capability inventory: Includes file moves (mv), file writes, and pattern searching (grep).
Sanitization: The skill does not perform explicit sanitization of the input text before processing.
Mitigation: The workflow is highly iterative and requires user approval for each major step (identifying headers, grouping topics, and finalizing summaries), which significantly mitigates the risk of an automated injection attack succeeding without human oversight.

transcript-curator