Completion-Truth Guard Starter

A small set of completion-truth guards we run in our own AI operation daily: checks that compare an AI agent's "done" claim against physical state — file mtimes, exit codes, diffs, HTTP responses — before the claim is accepted.What you get:- Three guards (completion receipt, overclaim checklist, session handoff) plus a Claude Code stop hook that detects fabricated tool results — designed for Claude Code / CLI-agent style workflows, with notes on where to adapt them to your setup- A setup walkthrough with real examples from our own operation, including the failures that made us build this- Two weeks of async Q&A (email or dev.to comments)Everything async, no meetings.Use freely in your own projects; please don't resell the pack itself.

Found on Devadex — the discovery index for independent software the big search engines bury. More from gumroad.

Report this listing