Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Full-page web reading for non-academic perspectives — blogs, tech reports, product pages, industry analysis. Spawns a subagent to read pages in isolated context. Hard constraint: at least 30 web pages read in full.
.claude/skills/yogsoth-ai-deep-web-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-20 | ✓→✓ | ＝ Same ✓ | 4% | 0% |
| case-21 | ✓→✓ | ＝ Same ✓ | 49% | 0% |
| case-22 | ✓→✓ | ＝ Same ✓ | 46% | 0% |
| case-06 | ✗→✗ | ＝ Same ✗ | -18% | 0% |
Deep web reading for non-academic perspectives (blogs, tech reports, products, industry analysis).
Subagent — spawned via subagent-spawning/spawn-agent skill. The subagent reads pages in its own context window, protecting the main session from context overflow.
At least 30 web pages read in full before completing this SOP.
Reading 30+ full web pages consumes significant context. Running this as a subagent isolates the heavy reading from the main dialogue session. The subagent returns a structured summary, not raw page content.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 25,203 | 23,413 | -7% | 1 | 1 | 0% | 3,769 | 3,020 | -20% | 0 | 0 | — |
case-06 | fail→fail | 39,948 | 35,294 | -12% | 1 | 1 | 0% | 6,179 | 5,083 | -18% | 0 | 0 | — |
case-01 | fail→fail | 39,581 | 38,815 | -2% | 1 | 1 | 0% | 6,187 | 6,213 | +0% | 0 | 0 | — |
case-02 | fail→fail | 16,995 | 20,305 | +19% | 1 | 1 | 0% | 2,554 | 813 | -68% | 0 | 0 | — |
case-03 | fail→fail | 37,793 | 47,513 | +26% | 1 | 1 | 0% | 6,184 | 6,405 | +4% | 0 | 0 | — |
case-04 | fail→fail | 14,946 | 17,724 | +19% | 1 | 1 | 0% | 1,926 | 2,061 | +7% | 0 | 0 | — |
case-05 | fail→fail | 32,336 | 38,132 | +18% | 1 | 1 | 0% | 4,856 | 4,785 | -1% | 0 | 0 | — |
case-08 | fail→fail | 17,813 | 35,364 | +99% | 1 | 1 | 0% | 2,784 | 5,656 | +103% | 0 | 0 | — |
case-09 | fail→fail | 21,707 | 17,647 | -19% | 1 | 1 | 0% | 3,177 | 699 | -78% | 0 | 0 | — |
case-10 | fail→fail | 27,387 | 15,501 | -43% | 1 | 1 | 0% | 4,084 | 996 | -76% | 0 | 0 | — |
case-11 | fail→fail | 17,731 | 8,356 | -53% | 1 | 1 | 0% | 2,637 | 659 | -75% | 0 | 0 | — |
case-12 | fail→fail | 20,133 | 11,232 | -44% | 1 | 1 | 0% | 3,055 | 751 | -75% | 0 | 0 | — |
case-13 | fail→fail | 16,800 | 46,378 | +176% | 1 | 1 | 0% | 2,413 | 6,377 | +164% | 0 | 0 | — |
case-14 | fail→fail | 19,486 | 40,356 | +107% | 1 | 1 | 0% | 3,037 | 6,389 | +110% | 0 | 0 | — |
case-15 | fail→fail | 24,176 | 16,001 | -34% | 1 | 1 | 0% | 3,687 | 1,028 | -72% | 0 | 0 | — |
case-16 | fail→fail | 18,157 | 13,696 | -25% | 1 | 1 | 0% | 2,706 | 743 | -73% | 0 | 0 | — |
case-17 | fail→fail | 17,460 | 44,944 | +157% | 1 | 1 | 0% | 2,655 | 6,254 | +136% | 0 | 0 | — |
case-18 | fail→fail | 18,376 | 16,384 | -11% | 1 | 1 | 0% | 2,712 | 787 | -71% | 0 | 0 | — |
case-19 | fail→fail | 25,476 | 14,732 | -42% | 1 | 1 | 0% | 3,951 | 823 | -79% | 0 | 0 | — |
case-20 | pass→pass | 37,560 | 38,456 | +2% | 1 | 1 | 0% | 6,187 | 6,408 | +4% | 0 | 0 | — |
case-21 | pass→pass | 5,984 | 7,827 | +31% | 1 | 1 | 0% | 1,039 | 1,546 | +49% | 0 | 0 | — |
case-22 | pass→pass | 24,184 | 38,934 | +61% | 1 | 1 | 0% | 4,280 | 6,247 | +46% | 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, and 12 counted toward the lift figure. The other 10 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +5 percentage points is the difference between those two pass rates over the 12 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.