harness-mcp
Warn
Audited by Snyk on Feb 27, 2026
Risk Level: MEDIUM
Full Analysis
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (high risk: 0.80). The skill explicitly ingests user-generated pull request activities via harness_get_pull_request_activities (see "Get PR Activities (Comments, Reviews)" and the "Sync PR Comments to Jira" example) and uses those comments/reviews to create Jira comments and drive PR-to-Jira transitions, meaning untrusted third-party PR/comment content is read and can materially influence actions.
MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).
- Potentially malicious external URL detected (high risk: 0.90). The MCP server runtime configuration runs remote code with "npx -y @anthropic-ai/mcp-harness" (i.e., fetching and executing the package from the npm registry: https://www.npmjs.com/package/@anthropic-ai/mcp-harness), which is a required runtime dependency that will execute remote code and can directly control agent behavior.
Audit Metadata