paragon-try-context

Pass

Audited by Gen Agent Trust Hub on Nov 2, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes curl for API interaction and openssl for local JWT generation. These commands are legitimate for a developer-focused integration tool and are executed within the user's terminal environment.
  • [DYNAMIC_EXECUTION]: The provided script scripts/paragon-env.sh sources a local configuration file at ~/.paragon/agent-os.env to manage project-specific environment variables. This is a standard and expected pattern for CLI tool configuration.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied project and user identifiers which are written to a shell environment file. The following evidence chain applies:
  • Ingestion points: The agent prompts for Project ID, Region, and User ID in SKILL.md to populate the configuration file.
  • Boundary markers: A quoted heredoc ('EOF') is used in the cat command to prevent shell expansion of the inputs during the file creation process.
  • Capability inventory: The skill uses openssl for cryptographic signing and curl for network requests to vendor-owned domains.
  • Sanitization: Input is not explicitly sanitized, but the workflow is local to the user's machine and includes explicit guidance on maintaining restrictive file permissions (chmod 600).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 2, 2026, 08:16 PM