The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it fetches transcription results from a remote service and presents the raw text directly to the agent's context.
Ingestion points: scripts/speech-to-text.sh (line 62) fetches content via curl from a RESULT_URL generated by the API.
Boundary markers: The transcription content is not wrapped in delimiters or accompanied by instructions to ignore embedded commands.
Capability inventory: The skill utilizes the Bash tool for script execution and the Write tool for configuration management.
Sanitization: No sanitization, escaping, or validation is performed on the fetched transcription text before it is displayed to the agent.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of remote resources for processing.
The deapi_ensure_local_file function in scripts/common.sh uses curl to download files from user-provided URLs passed via --audio or --ref-audio arguments to a temporary directory.
[COMMAND_EXECUTION]: The skill relies on shell scripts to perform its core operations, including network communication with the deAPI infrastructure and local file management.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a dynamic context injection pattern used for configuration verification.
Evidence: ! cat ${CLAUDE_SKILL_DIR}/config.json 2>/dev/null || echo "NOT_CONFIGURED" executes at skill load time to check for the existence of an API key.

deapi-audio