The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to perform file system maintenance tasks. This includes executing the date command to generate timestamps, and performing rm, find, and mv operations to delete processed text files and relocate associated audio files from ~/ASR and /tmp/audio.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its transcript processing workflow. 1. Ingestion points: The agent reads untrusted text data from files in the output/ directory (SKILL.md, Step 2). 2. Boundary markers: While the skill identifies specific sections like === TRANSCRIPTION ===, it lacks explicit delimiters or instructions to ignore any embedded directives within the ingested text. 3. Capability inventory: The skill has access to high-privilege tools including Bash, Read, and Write. 4. Sanitization: No sanitization or validation is performed on the transcript content before the agent processes it according to Step 4, which could allow malicious content to influence the agent's behavior.

transcript-cleanup