The loop without the agent: webhooks and a REST API do the boring wiring

Not every part of the loop needs an agent. Two plain integrations carry most of the automatic bookkeeping: a git webhook and a REST feedback endpoint.

Commits that file themselves

Point your GitHub or GitLab webhook at the workspace and any commit whose message mentions an idea — #123, UP-123, or "idea 123" — links itself to that idea's Commits tab on push. One commit can reference several ideas; each gets the link. Idempotent by SHA, so replays and force-pushes don't double-count.

The webhook also answers "who wrote this?" on its own. Commits carrying agent trailers — Co-Authored-By: Claude, GitHub Copilot, Cursor, Gemini, Devin, Aider and friends — or exact bot author identities are stamped AI-authored automatically, feeding the same AI-contribution reporting the MCP path uses. The matching is deliberately narrow: a human developer who happens to be named Claude stays human.

A feedback endpoint your support desk can hit

POST /api/feedback accepts a bearer API key and a JSON body — product, message, contact name and email, an optional summary and external reference. Contacts match by email so repeat reporters accumulate history; an externalRef makes retries idempotent (you get the original record back, not a duplicate); payloads are capped and requests are rate-limited to 60 per minute per key. This endpoint exists because a real integrator needed it — their support agent now files customer issues directly from their helpdesk, each one arriving pre-linked to a contact.

The quiet consequence

With commits and feedback wiring themselves, the workspace's reports stop being estimates. The delivery funnel, the per-owner scorecards, and the AI-contribution numbers are all downstream of these two dumb, reliable pipes — no ETL, no weekly export, no one "keeping the tracker updated".