The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted social media posts and interpolates them into LLM prompts for narration generation, creating a surface for indirect prompt injection. \n
Ingestion points: Data is fetched from the Bluesky API in scripts/ingest.ts and scripts/narrate.ts.\n
Boundary markers: Absent; the skill uses simple metadata labels like [ORIGINAL POST] but lacks robust delimiters or explicit instructions to ignore embedded commands.\n
Capability inventory: The skill performs subprocess calls to ffmpeg, remotion, and bun (scripts/render-video.ts, scripts/generate-video.ts), writes multiple JSON and media files to the workspace, and makes network requests to external APIs.\n
Sanitization: Absent; the social media post content is processed as raw text without validation or filtering for potentially malicious instructions.\n- [COMMAND_EXECUTION]: The skill utilizes child_process.spawn and Bun.spawn to orchestrate scripts, execute ffmpeg for media muxing, and trigger npx remotion for video rendering. While user-supplied handles are sanitized for file paths, the reliance on shell execution for orchestration remains a risk factor. Additionally, the Remotion configuration (remotion.config.ts) disables Chromium web security during the rendering process.\n- [EXTERNAL_DOWNLOADS]: The skill makes network requests to well-known service providers, including the Bluesky API (public.api.bsky.app) for data ingestion, the Anthropic API (api.anthropic.com) for narration generation, and the ElevenLabs API (api.elevenlabs.io) for voice synthesis.

tweetscape-narrator