create-agent-adapter

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSNO_CODE
Full Analysis
  • [COMMAND_EXECUTION]: The documentation describes an architecture that utilizes runChildProcess() to execute CLI-based agents on the host machine. This is the primary intended function of the adapter system. The guide includes safety warnings against using eval() on process output.\n- [EXTERNAL_DOWNLOADS]: The skill references several Node.js dependencies in its templates, including @paperclipai/adapter-utils and picocolors. These originate from the vendor or well-known public registries and are documented for development purposes.\n- [NO_CODE]: This skill consists entirely of technical documentation and code templates in Markdown format. It does not include any executable scripts or binaries within its own distribution.\n- [INDIRECT_PROMPT_INJECTION]: The guide describes a system that ingests and parses untrusted output from LLM-driven agents in server/parse.ts and ui/parse-stdout.ts. While this constitutes an attack surface, the documentation provides remediation guidance: \n
  • Ingestion points: Agent stdout and stderr parsed in the server and UI modules.\n
  • Boundary markers: The templates recommend structured parsing of output lines.\n
  • Capability inventory: Adapters are designed to use runChildProcess() for command execution and fetch() for network requests.\n
  • Sanitization: The guide explicitly recommends using safe extraction helpers (asString, parseJson) and warns against dynamic execution of agent output.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 07:41 PM