nudges: hints only push writing — add read-side nudges to tool outputs (the strongest proven discovery channel)

[1Croydan1]

Problem

The nudge economy is asymmetric: the product constantly pushes WRITING via tool outputs — and never READING. CAPTURE_NUDGE (kaeru-mcp/src/utils.rs:36-38) after every episode/cite/claim says "search related → link … chain(from,to)" — writing only. The only advertisement for read_chain lives in the output of chains, which itself is never called — a circular dependency.

Why this is the biggest lever

The audit proved that a hint in a tool's output is the strongest discovery channel after the user's personal CLAUDE.md:

• CAPTURE_NUDGE made link the fix: disable rmcp session keep-alive to prevent ✘ failed flicker #1 organic tool (751 calls);
• reflect's output ("chains gone stale (rechain)") produced rechain=3;
• open reviews in awake's output produced close_review=8.
  Meanwhile read-side tools without such a channel are dead: chains/read_chain/tagged/trace/surface — 0 calls ever; 141+ demotions into layer=cold and 0 reads via surface (archived knowledge is effectively unrecoverable).

Proposal — read-side nudges in the outputs of living tools

• search results footer: "↳ drill for neighbors, trace for provenance";
• node brief in drill/at/recall: "in chain: (step 3/7) — read_chain " (requires reading chain_member on the read path);
• layer cold/frozen output: "bring back into view: surface";
• claim output: "when the verdict lands — confirm/refute --by ; open ones: tagged "status:open"";
• search with 0 hits: suggest query* and tagged topic:…;
• awake/overview of a team initiative: "cloud: N shared, M not pulled — cloud_recall" (see the cloud issue).