security-scan

Warn

Audited by Gen Agent Trust Hub on Apr 14, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs users to install or run the 'ecc-agentshield' package from the NPM registry, which is an external source.
  • [COMMAND_EXECUTION]: Provides commands to execute the 'ecc-agentshield' tool via npx or global shell installation, granting external code access to the project's environment and file system.
  • [CREDENTIALS_UNSAFE]: Recommends users export their 'ANTHROPIC_API_KEY' to the environment to enable advanced analysis features, which exposes the sensitive key to the third-party scanning tool.
  • [DATA_EXFILTRATION]: The tool accesses the sensitive '.claude/' directory and mentions a '--stream' flag that could transmit project configuration details externally during AI-powered analysis.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it reads and analyzes files such as CLAUDE.md and agent definitions which could contain malicious instructions. \n
  • Ingestion points: The tool reads the contents of '.claude/settings.json', 'CLAUDE.md', and 'agents/*.md'. \n
  • Boundary markers: The skill does not specify any delimiters or safety markers to distinguish between data and instructions when reading scanned files. \n
  • Capability inventory: The 'ecc-agentshield' tool can execute shell commands, read and write local files, and potentially make network requests. \n
  • Sanitization: There is no evidence of input validation or content sanitization being performed on the scanned configuration files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 14, 2026, 12:20 AM
Security Audit — agent-trust-hub — security-scan