tonydzi /sostav-comments — draft replies to the fresh club shortlist (the owner's voice, draft-first)
Why. When the owner goes through the fresh nightly shortlist, he usually wants ready-made draft replies for the interesting club threads. This skill does it in one pass: read the detector report → pick the safe posts → write short answers in his voice → show them. It sends nothing — the owner copies whatever he likes.
The main rules:
- Draft-first, never publishes. The skill only writes drafts into the chat. Posting into the club is done by the owner, by hand.
- Safe-topic gate. Draft only in opinion/knowledge threads (Knowledge, Lectures, Business, Investing, Travel, Health, General-on-topic). ⛔ NEVER draft in the Crypto / Dating / Off-topic-flood threads, or in any grey-financial or personal thread — those are data, not a place to reply.
- Short text, tone set by the channel ([[short-text-when-unreviewed]]): a meaningful answer in messenger register (not a 5-word Facebook joke — here you need context and knowledge of who the person is; pull the
person-sostav-*card if the member is a ⭐). - The owner's voice (Opus). No "positioned" replies, no showing off. On topic, to the point, without making someone else's announcement about yourself.
- Anti-leak ([[reglament-anti-leak-na-vyhode]]): drafts stay in the private layer; nothing sensitive from the club leaks into other channels.
0. A fresh shortlist
The latest detector report:
ls $IMPORTS_ROOT/alpha\candidates\sostav-*-report.md # take the newest one by date
If there is none / it is stale — refresh the corpus first (the nightly path): python $IMPORTS_ROOT/sostav\nightly_run.py (idempotent, it backfills).
1. Pick the posts worth answering
From the shortlist keep only those where:
- the topic is safe (see the gate above);
- the post is an opinion/question/piece of knowledge where a meaningful answer belongs (not an announcement, not an intro, not a grey request);
- a ⭐ author → open
$OBSIDIAN_VAULT/07-People\person-sostav-<slug>.mdfor the "who is this" context.
2. A grounding snippet (when it helps)
If the answer benefits from specifics (a quote/fact/example), take a relevant chunk from the working corpus — the YouTube Data API (the only pool that still works; the social networks are behind login walls), via $IMPORTS_ROOT/sostav\daily_safe_fetch.py (safe-only gate). Lift 1-2 fragments verbatim; don't paraphrase.
3. The drafts (Opus, personal, all different)
For each selected post — a separate short answer in the owner's voice. Assemble the batch:
1. [Knowledge · Author X] post: "…the claim…"
→ draft: "…a meaningful, on-topic answer in the owner's voice…"
2. ...
Different texts, on topic, no template. Voice quality ≥ Opus ([[content-Mei-style]] — an influence, not a copy).
4. Hand it to the owner
Show the batch of drafts in the chat + note which ⭐ contacts you pulled cards for. Stop there — the owner sends them himself.
Stop valves
- ⛔ Post/send nothing — drafts into the chat only.
- ⛔ Don't draft in grey/personal/crypto threads (the safe gate).
- ⛔ Nothing from the club leaks into other channels ([[reglament-anti-leak-na-vyhode]]).
- Money/commitments/secrets inside a draft → cut them, that is Tier-2.
Related
- Detector / nightly path:
$IMPORTS_ROOT/sostav\nightly_run.py,sostav_alpha.py. - The reply corpus:
daily_safe_fetch.py(YouTube Data API,secrets/youtube.env). - Voice: [[short-text-when-unreviewed]], [[content-Mei-style]],
/speak-as. - Sibling skills:
/fb-reply(Facebook comments),/mine-channel,/alpha-judge. - Canon: memory
sostav-community-import, [[reglament-anti-leak-na-vyhode]].
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.