Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Research topics via web search, fetch content, and file to inbox with provenance metadata. The first step for acquiring external knowledge. Chains into /seed and /pipeline for full processing. Supports multiple search engines and content types. Triggers on: "learn", "research", "look up", "find out about"
.claude/skills/miosa-osa-learn/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 51% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 6% | 0% |
> Research topics via web search and ingest into the knowledge pipeline.
Acquire external knowledge on a topic. Search the web, evaluate results, fetch the most relevant content, convert it to markdown with provenance metadata, and hand it off to /seed for processing pipeline entry. This is how the knowledge base grows from external sources — every piece of acquired knowledge gets full provenance tracking.
bash# Research a topic /learn "enterprise AI pricing models 2026" # Research and auto-process through pipeline /learn "HIPAA compliance for AI agents" --pipeline # Research with specific sources /learn "Monte Carlo tree search" --sources arxiv,wikipedia # Research with result limit /learn "competitor analysis SaaS pricing" --max 5 # Research and seed without full pipeline /learn "Firecracker VM security model" --seed-only # View learning history /learn --history
| Flag | Type | Default | Description | |------|------|---------|-------------| | <query> | positional | required | Research query or topic | | --max | int | 5 | Maximum sources to fetch | | --sources | string] | all | Limit to specific sources: web, arxiv, wikipedia, github, news | | --depth | enum | standard | quick (titles + snippets), standard (fetch + summarize), deep (fetch + full pipeline) | | --pipeline | flag | false | Auto-run through /pipeline after seeding | | --seed-only | flag | false | Seed but don't run pipeline | | --tags | string] | auto | Topic tags to apply | | --reason | string | — | Why this research is needed (aids retrieval) | | --history | flag | false | View past research sessions | | --output-dir | path | inbox/learned/ | Where to write fetched content |
/seed for inbox entry. If --pipeline, chain directly into /pipeline.markdown## Research Report: "enterprise AI pricing models 2026" **Sources searched:** 12 | **Fetched:** 5 | **Seeded:** 5 **Total content:** ~14,200 words ### Sources Acquired | # | Title | Source | Date | Relevance | Status | |---|-------|--------|------|-----------|--------| | 1 | Enterprise AI Pricing: 2026 Benchmarks | Bessemer Venture Partners | 2026-02 | 0.94 | seeded | | 2 | Per-Seat vs. Usage-Based: The AI Pricing Debate | a16z | 2026-01 | 0.89 | seeded | | 3 | AI SaaS Pricing Survey (n=200) | OpenView Partners | 2025-12 | 0.85 | seeded | | 4 | How We Price Our AI Product | Anthropic Blog | 2026-03 | 0.78 | seeded | | 5 | The Death of Per-Seat Pricing | TechCrunch | 2026-02 | 0.72 | seeded | ### Key Findings (cross-source) 1. Per-seat pricing remains dominant for enterprise AI (68% of surveyed companies) 2. Usage-based pricing growing fastest (42% YoY adoption increase) 3. Median enterprise AI price: $1,800-2,500/seat/year 4. Hybrid models (base + usage) emerging as best practice ### Coverage Gaps - No data found on AI pricing for healthcare-specific verticals - Limited comparison with open-source alternatives - Suggested follow-up: `/learn "AI pricing healthcare vertical"` ### Seeded Items Files written to `inbox/learned/`: - `2026-03-20-bessemer-ai-pricing.md` - `2026-03-20-a16z-pricing-debate.md` - `2026-03-20-openview-pricing-survey.md` - `2026-03-20-anthropic-pricing.md` - `2026-03-20-techcrunch-pricing.md` Next step: `/pipeline inbox/learned/2026-03-20-*.md --batch` to process all.
/seed — Downstream handoff for each acquired source/pipeline — Optional full processing chain| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 73,906 | 54,871 | -26% | 1 | 1 | 0% | 7,084 | 6,120 | -14% | 0 | 0 | — |
case-02 | fail→fail | 7,898 | 16,569 | +110% | 1 | 1 | 0% | 294 | 3,851 | +1210% | 0 | 0 | — |
case-03 | fail→fail | 17,492 | 14,556 | -17% | 1 | 1 | 0% | 2,504 | 3,771 | +51% | 0 | 0 | — |
case-04 | fail→pass | 17,743 | 12,470 | -30% | 1 | 1 | 0% | 3,482 | 3,271 | -6% | 0 | 0 | — |
case-05 | fail→pass | 12,477 | 8,592 | -31% | 1 | 1 | 0% | 1,672 | 2,531 | +51% | 0 | 0 | — |
case-06 | pass→pass | 5,826 | 9,143 | +57% | 1 | 1 | 0% | 850 | 2,842 | +234% | 0 | 0 | — |
case-07 | fail→fail | 2,683 | 11,452 | +327% | 1 | 1 | 0% | 385 | 2,882 | +649% | 0 | 0 | — |
case-08 | fail→fail | 14,610 | 7,359 | -50% | 1 | 1 | 0% | 2,231 | 2,571 | +15% | 0 | 0 | — |
case-09 | fail→fail | 7,205 | 12,826 | +78% | 1 | 1 | 0% | 453 | 3,460 | +664% | 0 | 0 | — |
case-10 | fail→fail | 14,640 | 11,147 | -24% | 1 | 1 | 0% | 1,746 | 2,858 | +64% | 0 | 0 | — |
case-11 | fail→fail | 57,943 | 16,969 | -71% | 1 | 1 | 0% | 8,219 | 4,303 | -48% | 0 | 0 | — |
case-12 | fail→pass | 31,891 | 9,750 | -69% | 1 | 1 | 0% | 5,182 | 2,993 | -42% | 0 | 0 | — |
case-13 | fail→fail | 16,947 | 18,072 | +7% | 1 | 1 | 0% | 2,720 | 4,653 | +71% | 0 | 0 | — |
case-14 | pass→pass | 36,893 | 25,704 | -30% | 1 | 1 | 0% | 6,467 | 5,739 | -11% | 0 | 0 | — |
case-15 | fail→pass | 15,699 | 9,746 | -38% | 1 | 1 | 0% | 2,813 | 2,973 | +6% | 0 | 0 | — |
case-16 | fail→pass | 32,920 | 16,671 | -49% | 1 | 1 | 0% | 3,195 | 3,643 | +14% | 0 | 0 | — |
case-17 | fail→pass | 52,502 | 34,392 | -34% | 1 | 1 | 0% | 8,207 | 6,243 | -24% | 0 | 0 | — |
case-18 | fail→pass | 6,613 | 10,462 | +58% | 1 | 1 | 0% | 274 | 3,010 | +999% | 0 | 0 | — |
case-19 | fail→fail | 5,754 | 8,451 | +47% | 1 | 1 | 0% | 242 | 1,778 | +635% | 0 | 0 | — |
case-20 | pass→pass | 21,849 | 18,758 | -14% | 1 | 1 | 0% | 3,803 | 4,423 | +16% | 0 | 0 | — |
case-21 | fail→pass | 5,485 | 28,529 | +420% | 1 | 1 | 0% | 210 | 4,802 | +2187% | 0 | 0 | — |
case-22 | fail→pass | 18,843 | 9,347 | -50% | 1 | 1 | 0% | 3,089 | 3,041 | -2% | 0 | 0 | — |
case-23 | fail→fail | 13,789 | 6,062 | -56% | 1 | 1 | 0% | 1,922 | 1,681 | -13% | 0 | 0 | — |
case-24 | fail→pass | 7,361 | 12,454 | +69% | 1 | 1 | 0% | 399 | 3,402 | +753% | 0 | 0 | — |
case-25 | fail→pass | 18,764 | 10,957 | -42% | 1 | 1 | 0% | 2,924 | 3,229 | +10% | 0 | 0 | — |
case-26 | fail→pass | 33,670 | 7,615 | -77% | 1 | 1 | 0% | 4,540 | 2,620 | -42% | 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. 26 cases were attempted, and 19 counted toward the lift figure. The other 7 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 +50 percentage points is the difference between those two pass rates over the 19 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.