The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's installation instructions guide users to clone a repository from an external, unverified source (github.com/Jane-xiaoer/paper-collage-ad-codex.git) and execute multiple shell scripts (e.g., check-deps.sh, setup-indextts2-mlx.sh, prepare-indextts2-voice.sh) included in the repo. This facilitates the execution of unverified remote code on the host system.
[COMMAND_EXECUTION]: The Node.js scripts provided as examples (render-keyframe-video.mjs, narrate-indextts2.mjs, and mix-audio.mjs) use child_process.execSync to run shell commands constructed by direct string interpolation of data from JSON manifest files. For example, variables like line.text, line.emotion, and track.file are interpolated directly into shell strings without sanitization or escaping, which could allow for arbitrary command injection if the manifests contain malicious input.
[EXTERNAL_DOWNLOADS]: The setup-indextts2-mlx.sh script is documented as downloading machine learning models to the local directory ~/.local/share/paper-collage-ad/. These downloads are not verified for integrity or safety, representing a risk of fetching potentially harmful content from a remote server.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests untrusted data from various JSON manifests (storyboard, voice, audio-mix) and uses this data within powerful capabilities like shell command execution.
Ingestion points: manifest files in the manifests/ directory (e.g., storyboard.json, voice.indextts2.json).
Boundary markers: None identified in the provided instructions or scripts.
Capability inventory: Execution of shell commands via execSync for video rendering, voice synthesis, and audio mixing.
Sanitization: The skill lacks explicit sanitization or escaping of the external content before it is interpolated into shell commands.

paper-collage-ad-codex