/comments — comments under our content (show · reply · grow into a DM)

🧒 End the report to a non-technical owner with a short "In plain words" recap. (memory eli5-always)

One pass: show the comments and reply surgically where a reply is actually expected. It collects and displays; it sends replies only on the owner's explicit "+". A thin layer (AK-47) over the engines in $IMPORTS_ROOT/content-factory\registry\.

🔴 BATTLE LESSON 2026-07-15 (memory bulk-replies-cringe-one-offer-converts). The goal "zero unanswered" is CANCELLED: a batch of 16 replies to old comments was deleted by the owner in full ("my batch went out… I'm embarrassed myself"), comment auto-posting was turned off, and one of our accounts lost access to a third-party group. In the same night ONE public call-to-action post landed a partner within 30 hours. → Default = stay silent. Reply only if: it is addressed to us OR the thread is alive (younger than ~a week) OR a person is personally waiting for an answer. Older than that — skip, and that is NOT a debt. → Ceiling ≤3 replies per pass, never as a back-to-back series inside someone else's group: pace and volume read as a bot regardless of how good the text is. → Value is measured in people who moved, not in the unanswered counter.

Boundaries: this is about comments under OUR published posts (pubmetrics.db). The separate club corpus has its own skill (/sostav-comments). Facebook comments will be pulled in once the Graph token is live — for now the source is Telegram.

1. Fresh collection (0 tokens)

First refresh the database so you don't reply to stale material:

• python tg_comments_collect.py (idempotent; exit 3 = the sending rail is busy → say "retry later", don't crash).

2. Show the state

• python pub_metrics.py status — how many unanswered in total.
• python pub_comments_report.py — digest of the unanswered ones (spam bots from spam_authors.txt are already flagged 🤖 and filtered out).
• python pub_metrics.py top-commenters --n 15 — who comments most often.

Show it to the owner compactly: N unanswered per platform · top-5 commenters · link to the dashboard _Dashboards\Pub-Registry-Metrics.html.

3. If the owner asks for replies ("answer the comments" / "+")

Draft-first, sending by hand or on "+":

1. RECALL the author before every reply to a human: /find <name> (namesearch) + grep the vault + the Telegram profile (get_full_user by author_id from comments). Know who you are writing to.
2. Tone per channel ([[short-text-when-unreviewed]]): Telegram = a meaningful on-topic reply with context (NOT a 5-word Facebook joke). The owner's voice = the top-tier model.
3. What to skip: spam bots 🤖 (no reply); old threads where a reply is no longer appropriate (decide explicitly).
4. Show the batch as a table "comment → draft → which account sends it" and wait for "+". After sending — pub_metrics.py mark-replied --cid <id> (or the next tg_comments_collect.py will set replied by itself).

4. Top commenters → a public call in the group, a DM on top

An active commenter with no CRM card = a candidate. Create the card (CRM), run a RECALL. ⭐ The offer (testers/a call/early access) goes publicly into the group, not by DM — a DM only as a personal nudge AFTER the public call (canon: the house rulebook offer-to-active-people-publicly-not-by-dm, memory public-offer-in-group). Don't send a DM without a "+" ([[telegram-account-identities]]). Clan members are not leads.

Boundaries / safety

• READ by default. Mass back-dated replying = a hard stop: batch only, and only with a "+".
• Spam filter: add a new bot wave to spam_authors.txt (author_id # reason).
• Nothing private from the comments leaks into other channels.

Related

The morning auto-ping of unanswered items into the fleet log chat (05:35, task "Pub-Comments-Morning") and the nightly collection (03:40, collect_pub_metrics.cmd) — this skill is the manual, on-demand twin of those routines. Canon: 00-System\Pub-Metrics-Registry.md, memory content-pub-registry.

Like this skill? It is one of 100 in second-brain-starter-kit: the second brain we built for ourselves and run every day at Palo Alto AI Research Lab. Install the whole set with npx skills add tonydzi/second-brain-starter-kit. Everything is open source and free, so take what you need.

Flagships worth a look on their own: secondop-panel (a second opinion from a panel of external models), claude-memory-tidy (stop your agent's memory from rotting), telegram-mcp-kit (your own Telegram over MCP in about 15 minutes).

Author: Anton Dziatkovskii, Palo Alto AI Research Lab. Telegram @tonydzi - WhatsApp +1 341 222 9178 - X @Tony_Stef_

Engineers: want to test-drive this setup? Message me. I hand out free starter seeds to engineers who test and report back, and custom skill requests are welcome.