ericrisco/rsc-harness / skills/youtube-thumbnails
youtube-thumbnails
VerifiedUse when designing or fixing a YouTube thumbnail, making variants for an A/B test, or working out which thumbnail style wins on a channel — including the trap where CTR looks fine but the test keeps selecting the duller image on watch-time share. NOT pairing the title with the thumbnail as one click-promise (that is `youtube-packaging`), NOT channel positioning or niche (that is `youtube-strategy`), NOT setting a thumbnail through the Data API (that is `youtube-api`).
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 9.4 KB |
| evals/README.md | markdown | 720 B |
| evals/cases.yaml | config | 2.7 KB |
| references/composition-and-specs.md | markdown | 3.2 KB |
| references/experiment-log-format.md | markdown | 3.9 KB |
| scripts/verify.shexecutable | script | 3.4 KB |
references/experiment-log-format.md
View on GitHub# Experiment log format
The wiki layout, schema, filled example, and pattern-mining heuristic for the
feedback loop in SKILL.md. Everything lives under `02-DOCS/wiki/youtube/`.
`02-DOCS/wiki/` is an **OKF v0.1** bundle, so both files below carry file-level
YAML frontmatter with a non-empty `type` (`thumbnail-experiment`) plus the OKF
surface (`title`, `description`, `tags`, `timestamp`) — the same shared key set the
api/strategy/packaging siblings use. The frontmatter sits once at the top of each
file, above the table. Refresh `timestamp` when you append/rewrite. Any
cross-reference uses a standard markdown link (`[text](./file.md)`), never an
Obsidian `[[wikilink]]`.
## Files
- `thumbnail-experiments.md` — append-only log, one row per Test & Compare run.
- `thumbnail-patterns.md` — derived, human-maintained summary of what wins on this
channel. You rewrite the relevant line after each logged test.
## `thumbnail-experiments.md` schema
File-level frontmatter once at the top, then the append-only table below it:
```markdown
---
type: thumbnail-experiment
title: Thumbnail experiments log
description: One row per Test & Compare run — variants, axis tested, winner, and the takeaway.
tags: [youtube, thumbnails, ab-testing]
timestamp: 2026-06-01T00:00:00Z
---
# Thumbnail experiments
| date | video | variants | axis tested | winner | CTR | watch-time share | note |
```
- **date** — ISO date the test concluded.
- **video** — title or video ID.
- **variants** — short labels, e.g. `A: face-shock / B: object`.
- **axis tested** — the single axis you varied (subject, expression, text, color,
layout, zoom).
- **winner** — which variant YouTube auto-promoted.
- **CTR** — reported click-through for the winner (context only; not the deciding
metric).
- **watch-time share** — the metric that actually decided the test.
- **note** — one-line takeaway, especially any over-promise signal.
### Filled example
```text
| date | video | variants | axis tested | winner | CTR | watch-time share | note |
|------------|----------------------|---------------------------|-------------|--------|------|------------------|------|
| 2026-05-20 | I built a PC in 24h | A: face-shock / B: object | subject | A | 6.1% | 41% | face beat the build photo clearly |
| 2026-05-28 | $200 budget PC | A: shock / B: confident | expression | B | 5.4% | 39% | calm-confident edged shock for this niche |
| 2026-06-01 | Worst PC parts | A: text / B: no-text | text | A | 5.9% | 38% | one bold word helped; thin margin, re-test |
```
## `thumbnail-patterns.md` shape
```markdown
---
type: thumbnail-experiment
title: What wins on this channel
description: Derived summary of the settled thumbnail axes and their confidence.
tags: [youtube, thumbnails, patterns]
timestamp: 2026-06-01T00:00:00Z
---
# What wins on this channel
| axis | settled winner | evidence | confidence |
|------|----------------|----------|------------|
| subject | human face | 4/5 tests | high |
| expression | confident > shock | 2 tests | low |
| text | 1 bold word | 2 tests, thin margins | low |
```
## Pattern-mining heuristic
1. Group logged rows by `axis tested`.
2. For each axis, count which level won and by what watch-time margin.
3. Promote a level to "settled winner" only when it holds across ~3+ tests OR wins
by a large, repeated margin. Mark confidence honestly.
4. The next thumbnail keeps every settled winner and tests the next unsettled axis.
## Minimum-sample caveat
On a small channel, watch-time-share differences inside a single ~2-week test are
mostly noise. A single result is a hint, not a rule. Resist rewriting your whole
thumbnail style off one win — wait for the margin to repeat. When content type or
audience shifts noticeably, re-open settled axes; yesterday's winning pattern was
fit to yesterday's audience.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |