plan-topics
/plan-topics
Planuje tematy krótkich filmów na kolejny miesiąc: research, usuwanie duplikatów i niezależny przegląd podagenta przed zatwierdzeniem.
- Autor
- ishandas387 ↗
- Repozytorium
- ytshort-creator ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 12.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Planuje tematy krótkich filmów na kolejny miesiąc: research, usuwanie duplikatów i niezależny przegląd podagenta przed zatwierdzeniem.
Wyzwalacze z opisu
Oryginalny opis (EN)
Plan next month's Shorts topic file (or refill slots) with research, dedupe and an independent sub-agent review gate. Use for "plan next month", "refill topics", "replace slot N", or the monthly scheduled run.
Jak działa
- 1. Load
- 2. Slots
- 3. Draft (primary + reserve per slot). Every candidate must pass:
- 4. Independent review (mandatory)
- 5. Write
- 6. Close the log, then report
Podgląd SKILL.md
--- name: plan-topics description: Plan next month's Shorts topic file (or refill slots) with research, dedupe and an independent sub-agent review gate. Use for "plan next month", "refill topics", "replace slot N", or the monthly scheduled run. --- # Plan topics Run with logging: `run_id = YYYY-MM-DDTHHMMZ_topic-plan`, actor `session:plan-topics` (or `scheduled:topic-plan`), per `_system/LOGGING.md`. ## 1. Load Read `_system/logs/worklog.md`, `runs.csv`, the tail of `events.jsonl`, `_system/insights.md` (if it exists) and EVERY file in `_system/topics/`. Build: all used slugs (any status, including rejected), and each main subject with its last publish date. ## 2. Slots Every Mon/Wed/Fri/Sun date of the target month (the owner's timezone). The Wildcard (audience pick) is the last Sunday. Units = slots × 255. Above 4,500, warn and suggest a slot to drop. ## 3. Draft (primary + reserve per slot). Every candidate must pass: - **A. Story:** something happens: a person, ship, place or object, a moment of change, and a payoff in 25 seconds. Trivia comparisons are rejected. - **B. Hook:** 12 words or fewer, a concrete noun, ideally a number or a striking image. The stakes or contradiction is inside the hook. No meta openings, no "Did you know", no questions answerable with "no". Every word literally defensible. - **C. Visual:** one papercraft set with 3–5 beats of visible change (score 1–10). Avoid facial acting, documents, huge moving crowds, and abstract or map-only ideas. Write a one-line `visual_concept`. - **D. Facts:** open the sources. Tier A: museums, archives, academic, Britannica, Guinness, major news. Tier B: Wikipedia, History Today, Smithsonian, World History Encyclopedia. Never content farms, listicles or AI "fact" sites. Each core claim needs 1 Tier A source or 2 agreeing Tier B sources. For ranges, use the hedged value. Superlatives need a source that states them. Check for myth traps (later embellishments). - **E. Freshness:** search "<topic> shorts" and "<topic> tiktok". Saturated topics need a distinct angle and are capped at 1 in 4 slots. - **F. Sensitivity:** no events from the last 50 years with living victims, and no atrocities played for whimsy. Colonial or violent topics only with a sober angle, flagged `sensitive`. Hard rules: never reuse a slug. A reused subject needs a new angle, 90+ days' gap and a `sequel` note. No same pillar twice in a row. No three in a row sharing era or region. - Pillars: Ships & sea, Ancient world, Bizarre events, Lost & found, Disasters, War & peace, Wildcard. - Eras: Ancient (<500 AD), Medieval (500–1499), Early modern (1500–1799), 19th c, 20th c. - Regions: Europe, Middle East & N Africa, Sub-Saharan Africa, South Asia, East Asia, SE Asia & Oceania, Americas, At sea. Style: the house style from worklog/insights, with about 1 in 4 slots on the runner-up. Performance learning: use rows from the last 60 days that are 7+ days past publish and have metrics. With at least 6 such rows, steer and state the numbers; otherwise write "insufficient data". ## 4. Independent review (mandatory) Spawn a FRESH `topic-researcher` agent (Agent tool, `subagent_type: "ytshort-creator:topic-researcher"`) in **review mode**. Give it only the draft rows and "review mode"; the rules A–F, scoring and row verdict live in the topic-researcher agent definition. Never pass it your reasoning or research notes. (For research in step 3 you may also start `topic-researcher` instances in **research mode**, but never reuse a research instance to review.) Log each verdict (`subagent:topic-researcher`, `review.verdict`). A row passes only if ALL hold: fact PASS (or FIX applied); hook 7 or above; visual 6 or above; story Y; sensitivity OK (or a FLAG kept with a sober angle). Month level: average hook 7.5 or above, and saturated topics at most 1 in 4. On failure: - FIX: apply the wording. - FAIL or below threshold: swap in the reserve, then send only the changed rows to a FRESH reviewer. - After a maximum of 2 rounds, a row that still fails is written with status `needs_review`. - Rejected candidates go to the backlog as `rejected` with the reason. ## 5. Write `_system/topics/topics_YYYY-MM.csv` columns: `slot,publish_date,weekday,slug,title,status,style,pillar,era,region,key_entities,angle,sources,viewed_vs_swiped_pct,avg_pct_viewed,notes,hook,visual_concept,review` `sources` holds full URLs (opened, tiers marked), never site names alone; the script stage copies them into `script.md` and ytmanager into the description. The `hook` and `angle` columns are a draft. After an episode's script gate passes, `make-episode-hf` §2 overwrites them with the approved wording, so the CSV never disagrees with what ships. Then update the backlog: remove used rows, add rejected rows, and add 5–10 fresh unreviewed ideas, each with a source and a draft hook. ## 6. Close the log, then report Write the worklog entry (slots, units and headroom, review rounds and round-1 pass rate, rejections, flags, learning note) and the final `runs.csv` row. Then message the owner: a table of date | title | hook | review, plus a "Needs your decision" list. Spend no Krea units in this skill.
Wersje
W bazie nie ma innych skilli o nazwie plan-topics.
Migawka bazy z 13.11.2026 · wersja paczki e3d2e56c7071 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100yt-plan
Planuje tydzień lub miesiąc wgrywania filmów na YouTube: co publikować, kiedy i w jakiej kolejności. - 90google-ads-monthly-report
Przygotowuje miesięczny raport Google Ads dla klienta: leady z telefonów i formularzy, koszt leada i plan na kolejny miesiąc. - 100month-end-report-generator
Pisze miesięczny raport Google Ads, porównując go z poprzednim miesiącem, rokiem i celami, z oceną i planem na kolejny miesiąc. - 90marketing-growth
Planuje tematy, tytuły, miniatury i dystrybucję Shorts na kanale YouTube oraz analizuje dane twórców. - 100kai-monthly-audit
Tworzy miesięczny raport marketingowy dla kierownictwa z danymi z 30 dni, porównaniem z poprzednim okresem i planem na kolejny miesiąc. - 100kai-monthly-audit
Tworzy miesięczny raport marketingowy dla kierownictwa z danymi z 30 dni, porównaniem z poprzednim okresem i planem na kolejny miesiąc.