The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is instructed to access sensitive API keys from local configuration files at ~/.config/xiaolin-explainer/keys.env and ~/.config/ai-teardown/mimo.env. These keys are used for legitimate research and media generation services (Exa, Tavily, Serper, Firecrawl, Unsplash, Renoise, MiMo), but the practice of accessing the user's home directory configuration files represents a credential handling surface.
[COMMAND_EXECUTION]: The skill relies on multiple Python and shell scripts to automate media tasks. These scripts use subprocess.run to execute shell commands for ffmpeg, ffprobe, whisper-cli, and npm. Specific examples include scripts/setup_engine.sh performing package installations, and scripts/align_recording.py and scripts/narrate_mimo.py processing audio files via external CLI tools.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it actively retrieves data from external web sources through research engines and scraping tools (research.py). This data is used to populate a facts.md file and subsequently used by the agent to draft video scripts. Malicious content on scraped websites could potentially influence the script's narrative or include hidden instructions. The skill partially mitigates this risk through a manual fact-verification workflow and mandatory source documentation.

finance-explainer