Feature idea: agent-mediated feedback with curated aggregation visible back to users

The idea

Two connected pieces:

1. Let users file feedback through the agent, in-flow. Mid-session, the natural moment to report something is the moment it happens — "Claude, send that to Anthropic as feedback." Today the routes are /bug (human-initiated) or this repo (leaves the terminal, requires a GitHub account and the knowledge that this repo exists). The agent already has the full context of what went wrong; with explicit user consent per submission, it's the ideal filing clerk.

1. Close the loop with AI-curated aggregation. No human team can read all feedback — but a model can cluster it. Surface the clusters back: "your report joined ~400 similar ones about stale session dates; status: under consideration." Users who see their feedback land somewhere file more of it, and better. Users who file into a void stop.

Why it matters

The gap between "users who hit a paper-cut" and "users who report it" is enormous, and it's mostly friction and futility, not apathy. An agent that files consented feedback in-flow attacks the friction; a visible cluster/status attacks the futility. (I run exactly this pattern at small scale for my own product's launch feedback — feedback → AI curation → disposition visible to the person who gave it — and it measurably increases repeat feedback.)

Origin of this suggestion: I wanted to report the session-date-staleness issue (#73800) and realised I had no idea how to "get through" to Anthropic — the agent had to tell me this repo existed.

🤖 Filed with Claude Code