Video Researcher
Two research jobs for a video team. Mode 1 watches the market: every week it scores each video on a declared set of channels against that channel's own age-matched median, tests each theme against the rest of the channel's videos, and drops ranked idea cards on a Research board, each carrying the video, the metric and when it was measured. Mode 2 reads the product: it reads the released code, not the marketing page, and returns a fact sheet where every claim cites a file and a line. EN, DE and DA. Cron runs the scan, a heartbeat catches a missed week. Four verifiers print PASS or FAIL with numbers. guidance-only (nested skills:[{id,schema}] with discovery-only methods + peer_dependencies) — agents reach it via skill_read.
I built this after the first honest look at my own market gave me the wrong answer. The top of the outlier list said web design, with videos at 13 and 7 times their channel's normal views. Tested against the rest of each channel's videos, the lift was 0.99x once the one channel that made 41 of the 58 videos was removed. So this skill never ranks a video on raw views. It scores each one against the median of its own channel, in the same format and age band, then asks whether a theme survives losing its biggest channel. One theme did, at 1.68x. Every Monday a cron job runs the scan in English, German or Danish and drops ranked idea cards on a Research board, each with the video, the ratio and the date it was measured. You pick. Mode 2 reads a product's released code before a tutorial is scripted, and every claim cites a file and a line. Its first read contradicted my own docs. The scraping runs on your own scrapecreators access. Fact sheets feed @di-atomic/video-director and @di-atomic/copy-engine.
Features5
Score every video against its own channel's median, same format and age band.
Test each theme against the rest, then again without its biggest channel.
Drop ranked idea cards with the video, ratio and date on a board you pick from.
Read the released code and trace every tutorial claim to a file and a line.
Run the scan on a weekly cron, with a heartbeat that re-runs a missed week once.
Videos1 video
DemoVideo Researcher · Demo video
Demo · SpiderMediaArticles about Video Researcher2 articles
Stop guessing your next video. Your agent researches it every week
opvs.ai24 Sep 2026Stop guessing your next video. Your agent researches it every week
@di-atomic/video-researcher is live. Every Monday your agent scores competitor videos against their own channel, tests each theme, and drops ranked ideas with the evidence on a board you pick from.
Videos1 video
DemoVideo Researcher · Demo video
Demo · SpiderMediaArticles about Video Researcher2 articles
Showing all 2
SKILL.mdthe file your agent reads
SKILL.md loads when you open this tab.
What's new3 releases · newest first · from the package's CHANGELOG
v0.1.22026-11-10Latest
Marketplace listing update: added the launch explainer film (`media.demo_video_url`). No code, schema, copy or guidance changes; hero and screenshots unchanged from 0.1.1.
v0.1.12026-09-24
Propagation patch. No change to the method, the gates, the idea-card shape, the fact-sheet shape or the run contract. ### Added - **Marketplace card media.** `media.hero_image_url` plus four screenshots (the own-channel baseline, the peer chain, the theme test with the top channel removed, docs versus released code), rendered once and reused across the launch blog, both listing surfaces and the card. ### Changed - **Card copy rewritten to the card rules.** `marketing.long_description` is cut to under 200 words, with its numbers as numerals (0.99x, 1.68x, 41 of 58) and the peers named. Five feature descriptions, each 90 characters or fewer and verb-led. First person throughout. ---
v0.1.02026-09-24
First release. Guidance-only: nested `skills:[{id, schema}]` with six discovery-only methods and
four coordination peers.
### Added
- **Mode 1 — watch the market.**
- Scores videos against their own channel's median, split by format (long vs Shorts) and by age
band, with a floor of 8 videos per band.
- Themes are graded by lift against the rest, robustness with the top channel removed, breadth
and n.
- Ranked idea cards land on a Research board.
- **Mode 2 — read the product.** A fact sheet from the released code, with every claim traced to
`path:line@ref`.
- **Six recipes:** weekly-market-scan, one-off-channel-study, set-up-the-research-board,
schedule-the-scan, product-read, fact-check-a-script.
- **Four gates** that print PASS/FAIL with numbers: `verify-outlier-evidence`,
`verify-theme-lift`, `verify-claim-trace`, `lint-secrets`. Plus `parse-tolerant.mjs` for data
responses that contain raw control characters.
- **EN · DE · DA:**
- transcripts follow each video's own caption language;
- a small-count floor for small markets;
- machine-translated titles are never used alone for tagging.
- **Scheduling:** a weekly cron job, with a heartbeat watchdog that re-runs a missed or failed
week once.
- **22 registry items** (14 method assets, 8 dated worked examples) and **14 learnings**.
### Known limits
- YouTube only.
- The theme thresholds (2.0 / 1.5 / breadth 2 / n 20), the per-band floor of 8, and the
500-view small-count floor are judgement. They were calibrated on one scrape; recalibrate after
four weekly runs.
- Paid distribution of a video is not observable, and the skill never infers it.Showing all 3
