Surface early churn signals

Every day, do the following:
1. Call list_keywords to confirm "{PRODUCT_NAME}" is tracked in Octolens and get its keyword ID.
2. Call list_mentions to pull all mentions from the past 24 hours matching that keyword with sentiment: negative. Also call list_tags and pull mentions tagged with frustration- or churn-adjacent tags if the workspace has them (e.g. bug_report, complaint).
3. For each mention, classify churn risk into exactly one bucket:
- high — explicit switching language ("moving to", "cancelling", "looking for an alternative"), a blocker that stops their core workflow, or public frustration from an identifiable customer
- medium — a real complaint about a specific feature, pricing, or reliability, but no switching language
- skip — vague negativity, jokes, competitor fans, posts not actually about the product
4. Drop everything classified as "skip".
5. For each remaining mention, check whether it was already flagged: keep a running list (or search your ticketing tool if connected) for the source URL. If already flagged, skip it.
6. For each new high or medium signal, produce an entry:
**Risk:** high | medium
**Quote:** > {original quote, trim to ~300 chars}
**Source:** {url} ({source}, {timestamp})
**Author:** {authorName} (@{author})
**Why it matters:** one line on what's broken or at risk
**Suggested next step:** one line — e.g. "CS reach-out", "file bug", "reply publicly"
**Draft reply:** 2–3 sentences, helpful and human, no marketing tone. Acknowledge the specific problem. Never promise timelines.
7. Post the daily digest in chat, high risks first:
- Signals found: N (X high, Y medium)
- Already flagged: N
- The entries from step 6
If there are no new signals, say "No churn signals in the last 24 hours" and exit cleanly.
Do not contact anyone directly. Drafts are for a human to review and send.