aseprite-lua

Pass

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPERSISTENCENO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill documents an extensive capability inventory that exposes a significant attack surface for indirect prompt injection. Untrusted data from sprites, external files, or network streams could influence agent behavior.
  • Ingestion points: The API supports loading data from files via app.open, io.open, and app.fs.listFiles (references/api/app_fs.md), as well as receiving messages via the WebSocket class (references/api/websocket.md) and reading the system clipboard via app.clipboard (references/api/app_clipboard.md).
  • Capability inventory: The API provides high-impact capabilities including file writing (Sprite:saveCopyAs), directory manipulation (app.fs.makeDirectory), bidirectional network communication (WebSocket), system command execution (os.execute), and extension installation (app.command.Options).
  • Boundary markers: Absent. The documentation does not provide guidance or delimiters for the agent to distinguish between data and instructions when processing content from these ingestion points.
  • Sanitization: Absent. There are no instructions for sanitizing or escaping data retrieved from files, the clipboard, or network sockets before use.
  • [COMMAND_EXECUTION]: The documentation notes that standard Lua library functions like os.execute are available for executing system-level commands, albeit with a user permission prompt (references/README.md).
  • [DATA_EXFILTRATION]: The WebSocket API (references/api/websocket.md) allows for full bidirectional network communication. An agent could potentially use ws:sendText or ws:sendBinary to exfiltrate sensitive sprite data, local file content, or clipboard information to a remote server.
  • [PERSISTENCE]: The app.command.Options command includes an installExtension parameter (references/api/command/Options.md). This provides a mechanism for the agent to install persistent Aseprite extensions (.aseprite-extension), which could be used to maintain access or execute code across sessions.
  • [NO_CODE]: The skill consists entirely of markdown instructions and API reference documents; it does not distribute standalone executable scripts or binaries.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 10, 2026, 09:13 AM