paperbanana

Pass

Audited by Gen Agent Trust Hub on Mar 7, 2026

Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The connectivity probe script scripts/probe_openrouter_models.py prints the first 6 and last 4 characters of the OPENROUTER_API_KEY to the standard output, risking partial credential exposure in logs.- [COMMAND_EXECUTION]: The skill uses the paperbanana and paperbanana-mcp command-line tools to execute generation and evaluation tasks, which involve local file system access and subprocess management.- [EXTERNAL_DOWNLOADS]: The skill makes network requests to the OpenRouter API (openrouter.ai) for model verification and inference, which is a well-known technology service.- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection when processing methodology or data files.
  • Ingestion points: The skill reads content from methodology text files and statistical data files (CSV/JSON) provided by the user.
  • Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the processed files.
  • Capability inventory: The skill possesses capabilities to execute commands and communicate with network APIs based on instructions influenced by the input data.
  • Sanitization: No evidence of sanitization or validation of the input file content is present in the provided scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 7, 2026, 04:21 PM