"author:tsyyan" matching MCP servers:

Checks claims in AI agent reports — quotations, numbers, repo anchors, PR/CI status, and 'I ran it' outputs — against primary evidence such as byte snapshots, repository bytes, the GitHub API and session transcripts, with no model in the loop. Every check returns a re-runnable receipt and a non-zero exit code on failure, so a wrong quote or invented test result can be blocked at any hand-off.