The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill interacts with the ElevenLabs API (api.elevenlabs.io) to synthesize speech and check usage quotas. This is an official service for the skill's primary purpose and is recognized as a well-known service.
[COMMAND_EXECUTION]: The skill utilizes local system utilities such as say, afplay, ffplay, and ffmpeg to generate and play audio. These operations are performed via standard subprocess calls to handle audio synthesis and playback.
[DATA_EXPOSURE]: The skill accesses API keys from ~/.elevenlabs/api_key and ~/broomva/.env.local. This is a standard method for resolving credentials for the ElevenLabs service and is consistent with the skill's intended functionality.
[INDIRECT_PROMPT_INJECTION]: The talkback-hook.py script processes project transcripts to generate audio summaries. While this involves ingesting external data, the content is only used for text-to-speech synthesis.
Ingestion points: Reads Claude Code JSONL transcripts via a path provided on stdin in talkback-hook.py.
Boundary markers: None present; the script extracts text between  markers if found, otherwise uses the first few sentences.
Capability inventory: Performs file writes (audio files), local command execution (audio players), and network requests (TTS API).
Sanitization: The speakify function in talkback.py cleans markdown syntax, collapsing code blocks and URLs into spoken placeholders to reduce noise during playback.

talkback