The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process raw transcription data which constitutes untrusted input. It significantly reduces this risk by establishing a 'Triple Confirmation' protocol (checking against a dictionary, evaluating context, and finding evidence of drift) before making any modifications, ensuring the agent remains focused on the correction task rather than executing embedded commands.
[COMMAND_EXECUTION]: The skill utilizes standard file-access tools (Read, Edit, Write) to automate the transcription correction workflow. These operations are transparently documented and scoped to local archive directories and source file mirrors, which is appropriate for its stated purpose.
[CREDENTIALS_UNSAFE]: The documentation follows security best practices by guiding users to manage personal or sensitive terminology through local configuration files (user_dictionary.yaml) while explicitly instructing them to keep these files out of version control via .gitignore to prevent accidental credential or data leaks.

transcription-corrector