The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it scans the target app's source code to propose narratives and write bespoke scene files. Maliciously crafted comments or code within the target project could potentially influence the agent's instructions during the creative direction or code generation phases.\n
Ingestion points: Target app source code and .appshot-context.json (Phase 1).\n
Boundary markers: Absent. The skill does not define clear delimiters or instructions to ignore embedded commands within the scanned code.\n
Capability inventory: File system writes (Phase 3), shell command execution (npm, npx, kill), and local server hosting (Phase 4).\n
Sanitization: Absent. There is no evidence of validation or escaping for the content extracted from the project files before it is used in prompt construction.\n- [REMOTE_CODE_EXECUTION]: The skill generates custom .tsx scene files and an orchestrator script which are then executed at runtime via Remotion. Although this is the primary purpose of the skill, the execution of dynamically generated code from untrusted project scans represents a security risk.\n- [COMMAND_EXECUTION]: The skill utilizes several shell commands to manage the development and rendering environment:\n
Executes npm run dev to start a local preview server and uses kill or lsof to manage the process PID (Phase 4).\n
Runs npx remotion render to generate the final video files (Phase 5).\n- [EXTERNAL_DOWNLOADS]: The skill performs an npm install within the target project's directory to set up the Remotion environment and dependencies (Phase 3), which involves fetching packages from the npm registry.

appshot-videos