Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Outline documents before drafting them — the argument skeleton that gets alignment cheaply, the one-line-per-section discipline, and the review-the-outline step that saves rewriting the prose. Use when asked help me start this document, outline before I write, why do my docs get rewritten from scratch in review, or get sign-off before drafting. Produces the outline with each section's claim (not topic), the reader-and-decision header, the outline review step, and the expansion rules.
.claude/skills/mohitagw15856-outline-before-prose/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 58% | 0% |
Documents get rewritten in review because the structure was wrong — and structure feedback arrived after two days of prose polishing. The outline-first discipline moves that feedback to hour one: a skeleton where every section is a one-line claim (not a topic), a header naming the reader and the decision, and a five-minute outline review with the key stakeholder before any paragraphs exist. Prose expands an approved outline in one pass; prose written first defends itself against restructuring forever.
Ask for these if not provided:
Ordered as the argument · cuts noted: "considered and dropped: X (fails the header test)"]
Who · the one question · scheduled before drafting]
Fill-don't-spawn · the renegotiation trigger · the outline's afterlife as TOC/summary]
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,131 | 11,008 | -27% | 1 | 1 | 0% | 2,816 | 3,087 | +10% | 0 | 0 | — |
case-02 | fail→pass | 12,919 | 11,865 | -8% | 1 | 1 | 0% | 2,369 | 3,177 | +34% | 0 | 0 | — |
case-03 | fail→fail | 15,999 | 9,947 | -38% | 1 | 1 | 0% | 3,159 | 2,753 | -13% | 0 | 0 | — |
case-04 | fail→pass | 16,619 | 12,177 | -27% | 1 | 1 | 0% | 3,183 | 3,103 | -3% | 0 | 0 | — |
case-05 | fail→fail | 10,213 | 8,974 | -12% | 1 | 1 | 0% | 1,793 | 2,485 | +39% | 0 | 0 | — |
case-06 | pass→pass | 15,543 | 10,556 | -32% | 1 | 1 | 0% | 2,698 | 2,902 | +8% | 0 | 0 | — |
case-07 | fail→fail | 13,984 | 11,704 | -16% | 1 | 1 | 0% | 2,457 | 2,915 | +19% | 0 | 0 | — |
case-08 | fail→pass | 16,927 | 14,259 | -16% | 1 | 1 | 0% | 2,738 | 3,223 | +18% | 0 | 0 | — |
case-09 | fail→pass | 9,440 | 10,805 | +14% | 1 | 1 | 0% | 1,731 | 2,727 | +58% | 0 | 0 | — |
case-10 | fail→pass | 12,285 | 11,264 | -8% | 1 | 1 | 0% | 2,103 | 2,817 | +34% | 0 | 0 | — |
case-11 | fail→pass | 20,196 | 11,511 | -43% | 1 | 1 | 0% | 3,417 | 2,881 | -16% | 0 | 0 | — |
case-12 | fail→pass | 8,699 | 10,319 | +19% | 1 | 1 | 0% | 1,564 | 2,777 | +78% | 0 | 0 | — |
case-13 | fail→pass | 15,220 | 10,375 | -32% | 1 | 1 | 0% | 2,332 | 2,721 | +17% | 0 | 0 | — |
case-14 | fail→pass | 16,643 | 14,111 | -15% | 1 | 1 | 0% | 2,624 | 3,154 | +20% | 0 | 0 | — |
case-15 | fail→pass | 13,724 | 10,706 | -22% | 1 | 1 | 0% | 2,079 | 2,624 | +26% | 0 | 0 | — |
case-16 | fail→pass | 10,694 | 10,353 | -3% | 1 | 1 | 0% | 1,784 | 2,523 | +41% | 0 | 0 | — |
case-17 | fail→fail | 21,018 | 14,469 | -31% | 1 | 1 | 0% | 3,482 | 3,153 | -9% | 0 | 0 | — |
case-18 | pass→pass | 13,729 | 11,167 | -19% | 1 | 1 | 0% | 2,123 | 2,658 | +25% | 0 | 0 | — |
case-19 | pass→pass | 17,721 | 13,488 | -24% | 1 | 1 | 0% | 2,947 | 3,064 | +4% | 0 | 0 | — |
case-20 | fail→fail | 3,181 | 12,342 | +288% | 1 | 1 | 0% | 519 | 2,872 | +453% | 0 | 0 | — |
case-21 | fail→fail | 2,015 | 5,829 | +189% | 1 | 1 | 0% | 273 | 1,893 | +593% | 0 | 0 | — |
case-22 | fail→fail | 1,509 | 3,258 | +116% | 1 | 1 | 0% | 206 | 1,505 | +631% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +55 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.