# Where your buyers already talk, with a draft reply > Where your buyers already comment, with a draft reply. Drafts only, never sends. Loop: `engagement-mining` Category: Demand Autonomy: assisted License: MIT Source: https://vibemarketers.ai/loops/engagement-mining Finds public threads where your buyers are already discussing the problem, and drafts one reply each. Drafts only, and the loop has no send path. ## What you need - none required, a public search API is enough ## Goal prompt Give this to your agent. Replace anything in with your own. ``` Search public discussion surfaces for threads where is already discussing . Keep only threads with real discussion. For each, draft one reply that answers what was actually asked and would be useful to someone who never clicks anything. Write every draft to a drafts directory. Never post, never send, never authenticate to any surface that can publish. ``` ## Verifier prompt Run this as a second, separate agent. It grades the first one's work and must be able to fail. A checker that never fails is not a checker. ``` Read every draft and the loop's own code. FAIL if any send, post, publish or authenticate path exists anywhere in the loop. FAIL if any draft reads as promotion rather than an answer, if any draft contains a link that was not asked for, if a draft was written for a thread with no discussion, or if any draft left the drafts directory. Otherwise PASS. ``` ## Stopping condition Every surfaced thread has exactly one draft in the drafts directory, no send path exists anywhere in the loop, and the verifier marks PASS. ## How to run it 1. Save this file into your project, for example `.claude/loops/engagement-mining.md`. 2. Open your agent in that project (Claude Code, Cursor, or Codex). 3. Paste the goal prompt, with your own values in place of the angle brackets. 4. When it stops, start a fresh session and paste the verifier prompt. 5. Read the verdict before you accept any change. ## What happened when we ran it Last run 2026-09-28 on vibemarketers.ai. - Baseline: No process for finding where the ICP already discusses agent loops in public. The conversations were happening and nobody on this side was reading them. - Result: 4 queries run, 5 threads with discussion surfaced, 5 drafts written, 0 sends. The largest thread carried 320 comments. Each draft was checked against the rule that it must be useful to someone who never clicks anything. - What this does not prove: Nothing here shows a draft would be well received, because none was posted. Thread comment counts measure discussion, not the presence of buyers: a busy thread may be entirely engineers with no purchasing interest. - Cost: 0 human touches, 8 minutes, $0.00 - Raw log: https://github.com/genwfurukawa/vibemarketers-v2/blob/main/lab/runs/2026-09-28-engagement-mining.json ## Inside a company - Problem it solves: The buyers are already discussing the problem in public and nobody on the team is in those conversations. - Source of truth it reads: Public discussion surfaces and the Brain, so drafts answer from the same positioning as everything else. - What it may do: Read public threads, write drafts to a drafts directory. Never post, never send, never authenticate anywhere. - Who approves: A human reads every draft and posts it under their own name, or does not. - Who owns it: Whoever is willing to show up in public under their own name. - How you measure it: Drafts posted, and replies that got a response rather than silence. MIT licensed. Copy it, change it, run it. No attribution required.