projects

Warn

Audited by Gen Agent Trust Hub on Jun 12, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructs the agent to automatically search for and read API keys from sensitive local file paths. Specifically, it directs the agent to check ~/.claude/config.json for keys stored in the LaunchDarkly MCP server configuration, and to scan local .env files. Accessing a global agent configuration file can expose credentials for other unrelated services.
  • [COMMAND_EXECUTION]: The skill provides extensive templates and code examples for generating and executing local scripts across multiple languages (Python, Node.js, Go). It also includes instructions for running infrastructure-as-code (IaC) tools like Terraform, Pulumi, and Ansible, as well as command-line utilities for project management and verification.
  • [PROMPT_INJECTION]: The 'Explore the Codebase' workflow involves the agent reading and analyzing arbitrary files from the user's project. This creates an indirect prompt injection attack surface where a malicious project could include hidden instructions in its source code or documentation to influence the agent's behavior during the setup process.
  • [DATA_EXFILTRATION]: Performs legitimate network operations to the official LaunchDarkly API at app.launchdarkly.com to manage project configurations and environments.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 12, 2026, 08:05 PM
Security Audit — agent-trust-hub — projects