The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill rewrites user-supplied text into Civil War-era dispatches using an LLM. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the input data.
Ingestion points: Processes arbitrary user text provided via CLI arguments or file inputs in bin/slowburns.mjs and scripts/letterify.mjs.
Boundary markers: The system prompt in scripts/letterify.mjs uses --- delimiters to isolate user content and provides strict instructions to prevent the model from inventing facts or deviating from the intent.
Capability inventory: The skill possesses the ability to write files (audio, props, locks) and execute local commands (ffprobe, npm, node) which could be influenced by injected content if it were to bypass the initial LLM layer.
Sanitization: LLM output is trimmed and stripped of markdown formatting before being passed to the narration and rendering stages.- [COMMAND_EXECUTION]: The skill uses execFileSync in several scripts (build.mjs, narrate.mjs, letterify.mjs) to orchestrate the video production pipeline, including calls to ffprobe for audio validation and op for secure secret retrieval. Inputs to these commands are generally sourced from internal configurations or controlled file paths.- [EXTERNAL_DOWNLOADS]: The skill interacts with well-known external services for AI processing, including OpenRouter, ElevenLabs, and Cartesia. Additionally, it triggers the installation of Node.js dependencies for the Remotion framework on the first execution of scripts/build.mjs.

civilwar-letterifier