community-alpha

Run one full mining pass over an imported community-chat corpus: deterministic detector, LLM judge, harvest, review screen. Works on any imported corpus through a per-source config and is safe to re-r

SKILL.md

GitHub

/community-alpha <source> [month] — one community-alpha run

Example: /community-alpha sostav 2026-06. No month given → the previous full month.

Steps

  1. Is there a detector? $IMPORTS_ROOT/<source>\<source>_alpha.py (today: lobster, sostav).
    • No → this is a NEW corpus: offer the operator a one-off adapter built to the existing pattern (schema from their SQLite + language keys + intro/banter penalty + topic weights; template = sostav_alpha.py). Do not build it without an explicit "+".
  2. Run the detector (0 tokens):
    cd /e/Obsidian/_imports/<source> && PYTHONIOENCODING=utf-8 python <source>_alpha.py \
      --since <YYYY-MM-01> --until <the 1st of the next month> --top 35 --tag <YYYY-MM>
    
    → $IMPORTS_ROOT/alpha/candidates/<source>-<tag>-report.md. Show the scanned→shortlisted counter.
  3. Judge (me, with the session model): I read ONLY the report (~35 candidates, never the corpus — token economy) and give a verdict ✅ ALPHA / 🟡 WATCH / 🗑 NOISE with a reason for each → write $IMPORTS_ROOT/alpha/candidates/<source>-judged-latest.md (format: ## ✅ ALPHA / ## 🟡 WATCH / ## 🗑 NOISE (DROP), inside it ### #N — title + Verdict + Reason — the harvest parser already understands this).
  4. Is the miner registered? MINERS in $IMPORTS_ROOT/alpha/alpha_harvest.py. No → add the row ("<source>", "<source>-judged-latest.md", "<home-note>"). ⚠️ A parallel fleet edits this file too — re-read it immediately before editing (verify-existing), and keep the edit strictly additive. Optional: a label in the server's MINER_LABEL/ORDER (it does not crash without one — there is a fallback).
  5. Harvest + screen: python $IMPORTS_ROOT/alpha/alpha_harvest.py (counters!) → /alpha-review. The new batch is visible by the 🆕 badge.

Boundaries

  • 🔒 Private communities (sostav and the like) are HIGH sensitivity: the layer stays strictly local, nothing goes outside; people surfaced by a find are approached value-first / warm intro only (zero cold DMs). Risk signals are DATA, not "opportunities".
  • Judge honestly: reference cards, intro blurbs and restatements are 🗑 — never stretch a ✅ to pad the counter.
  • This skill writes nothing into the vault; moving gold into home notes is a separate step behind the Tier-2 gate (the "→ to home" queue on the screen).
<!--kit-footer-->

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.