Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →AI agent with retrieval tool for document Q&A using RAG and LangGraph.
.claude/skills/majiayu000-langchain-retrieval-agent/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -56% | 0% |
An AI agent with retrieval tool for document Q&A using RAG, powered by LangGraph.
bashgit clone --depth 1 https://github.com/Eng0AI/langchain-retrieval-agent.git .
If the directory is not empty:
bashgit clone --depth 1 https://github.com/Eng0AI/langchain-retrieval-agent.git _temp_template mv _temp_template/* _temp_template/.* . 2>/dev/null || true rm -rf _temp_template
bashrm -rf .git git init
bashpnpm install
SUPABASE_URL - Supabase project URLSUPABASE_PRIVATE_KEY - Supabase service role keyOPENAI_API_KEY - For embeddings and LLMbashpnpm build
bashpnpm dev
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | pass→pass | 3,297 | 1,782 | -46% | 1 | 1 | 0% | 499 | 496 | -1% | 0 | 0 | — |
case-22 | pass→pass | 14,892 | 14,440 | -3% | 1 | 1 | 0% | 3,205 | 3,214 | +0% | 0 | 0 | — |
case-23 | pass→pass | 16,027 | 11,294 | -30% | 1 | 1 | 0% | 2,658 | 2,626 | -1% | 0 | 0 | — |
case-01 | fail→pass | 18,379 | 8,643 | -53% | 1 | 1 | 0% | 1,909 | 1,000 | -48% | 0 | 0 | — |
case-02 | fail→pass | 16,354 | 5,343 | -67% | 1 | 1 | 0% | 2,205 | 1,496 | -32% | 0 | 0 | — |
case-03 | fail→pass | 18,786 | 11,527 | -39% | 1 | 1 | 0% | 2,951 | 1,313 | -56% | 0 | 0 | — |
case-04 | fail→pass | 8,218 | 3,594 | -56% | 1 | 1 | 0% | 1,434 | 864 | -40% | 0 | 0 | — |
case-05 | pass→pass | 11,048 | 2,197 | -80% | 1 | 1 | 0% | 1,507 | 733 | -51% | 0 | 0 | — |
case-06 | pass→pass | 12,743 | 8,137 | -36% | 1 | 1 | 0% | 1,357 | 842 | -38% | 0 | 0 | — |
case-07 | pass→pass | 8,303 | 7,178 | -14% | 1 | 1 | 0% | 1,476 | 615 | -58% | 0 | 0 | — |
case-08 | fail→pass | 9,034 | 3,204 | -65% | 1 | 1 | 0% | 1,639 | 713 | -56% | 0 | 0 | — |
case-09 | pass→pass | 6,590 | 1,872 | -72% | 1 | 1 | 0% | 915 | 571 | -38% | 0 | 0 | — |
case-10 | pass→pass | 12,705 | 2,351 | -81% | 1 | 1 | 0% | 1,851 | 576 | -69% | 0 | 0 | — |
case-11 | pass→pass | 2,776 | 1,967 | -29% | 1 | 1 | 0% | 445 | 587 | +32% | 0 | 0 | — |
case-12 | fail→pass | 7,422 | 1,534 | -79% | 1 | 1 | 0% | 1,293 | 485 | -62% | 0 | 0 | — |
case-14 | pass→pass | 5,305 | 1,565 | -70% | 1 | 1 | 0% | 793 | 517 | -35% | 0 | 0 | — |
case-15 | pass→pass | 5,356 | 2,567 | -52% | 1 | 1 | 0% | 763 | 644 | -16% | 0 | 0 | — |
case-16 | pass→pass | 6,680 | 2,284 | -66% | 1 | 1 | 0% | 874 | 704 | -19% | 0 | 0 | — |
case-17 | fail→pass | 10,290 | 2,199 | -79% | 1 | 1 | 0% | 1,745 | 554 | -68% | 0 | 0 | — |
case-18 | fail→pass | 14,893 | 3,361 | -77% | 1 | 1 | 0% | 2,084 | 816 | -61% | 0 | 0 | — |
case-19 | fail→pass | 3,465 | 2,271 | -34% | 1 | 1 | 0% | 563 | 609 | +8% | 0 | 0 | — |
case-20 | pass→pass | 16,648 | 9,700 | -42% | 1 | 1 | 0% | 2,383 | 2,150 | -10% | 0 | 0 | — |
case-21 | pass→pass | 7,776 | 6,716 | -14% | 1 | 1 | 0% | 1,534 | 1,609 | +5% | 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. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 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.