toon-format

Pass

Audited by Gen Agent Trust Hub on Mar 18, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation includes installation instructions for the @toon-format/toon package via NPM, which is the core library for the data format described. It also references well-known packages such as openai, @anthropic-ai/sdk, gpt-tokenizer, and express for its implementation examples.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its core function of processing external data for LLM prompts.\n
  • Ingestion points: The CLI tool, decode function, and Express middleware examples all ingest arbitrary TOON-formatted strings from files, standard input, or network request bodies.\n
  • Boundary markers: Documentation examples suggest using Markdown code blocks (e.g., ```toon`) as delimiters when injecting data into prompts, but no programmatic enforcement of boundaries or "ignore instructions" directives are provided in the library itself.\n
  • Capability inventory: The provided code samples demonstrate the ability to read and write to the local file system (fs.promises) and interact with external LLM APIs (OpenAI and Anthropic).\n
  • Sanitization: The encoding and decoding logic does not include sanitization or filtering to detect or neutralize embedded LLM instructions (e.g., jailbreak attempts) within the data being processed.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 18, 2026, 11:44 PM