I tried having my agents read the vendor s docs before jumping into tasks, but in a lot of cases the docs are simply incomplete or stale. I burn tokens working around that — and the maintainer never finds out their docs are broken. The signal, an AI just hit a wrong page in your docs, gets dropped on the floor.FixYourDocs is an attempt to give that signal somewhere to go. It s an open protocol (one JSON shape, one POST /v1/reports) plus a hub that routes each report to whoever has verified they own the doc — by domain or by GitHub Pages repo — and opens a GitHub Issue on their repo. There are two SDKs (pip install fixyourdocs, npm i @fixyourdocs/sdk) and an MCP server (npx -y @fixyourdocs/mcp-server) that exposes a single file_doc_feedback tool.Try it yourself: ``` curl -X POST https://hub.fixyourdocs.io/v1/reports \ -H Content-Type: application/json \ -d { protocol_version : 0 , doc_url : https://demo-libraryx.com/payments , agent : { name : claude-code }, report : { kind : outdated , summary : Payments guide still shows the v1 charge flow (change me to avoid dedup) } } ``` and watch it land here: https://github.com/fyd-demo-maintainer/demo-libraryx/issuesHeads-up: this example opens a real, public GitHub Issue on the demo repo, so don t put anything you wouldn t want public in the summary.The landing page has a short demo that walks the whole loop — from an agent hitting a stale doc to the GitHub Issue that opens for the maintainer: https://fixyourdocs.ioIt s built with Claude and open: the protocol and SDKs are Apache-2.0, the spec text is CC-BY, all at docsfeedback.org. The hosted hub is FSL, converting to Apache-2.0 in two years.Landing page: https://fixyourdocs.io Spec: https://docsfeedback.org Code: https://github.com/fixyourdocsHappy to go deep on any of the details.