Add yarran-eng/literature-deep-reading

agents/yarran-eng__literature-deep-reading/README.md

@@ -0,0 +1,36 @@
+# Literature Deep Reading
+
+A universal, high-quality AI agent engineered for **intensive academic paper reading and research mentoring**. It transforms any LLM into a rigorous senior academic mentor and peer reviewer, guiding researchers through the true intellectual depth of any paper.
+
+## What it does
+
+Directly addresses the common pain point where generic LLMs mechanically paraphrase paper abstracts without genuine comprehension. Instead, this agent applies the **Feynman standard**: you have truly learned a paper when you can explain its core mechanism and key equations to a peer *without* looking at it.
+
+## Key Capabilities
+
+- **3-Stage State Machine** — analysis unfolds in structured stages, never all at once:
+  - **Stage 1 — Macro Skeleton**: Research question, prerequisite knowledge gates, contribution vs. baselines (structured comparison table)
+  - **Stage 2 — Mechanism Deep Dive**: Experimental design audit, figure/table decryption, ablation study analysis, maths intuition (intuition → formulation → stress test)
+  - **Stage 3 — Knowledge Internalization & Reproduction**: Feynman self-test questions, hostile reviewer critique of unstated limitations, transferable modeling tricks, step-by-step code reproduction roadmap
+
+- **Zero-Hallucination Epistemic Rigour** — always distinguishes: explicit author claims | reasonable inferences | speculative assertions
+- **Bilingual** — Chinese by default, English on request; enforces mandatory bilingual terminology anchoring
+- **Evidence Anchoring** — cites section numbers, figure/table IDs, dataset names, metric values, and hyperparameter settings
+- **No Fabrication** — never invents details absent from the paper
+
+## Example Usage
+
+1. Upload or paste an academic paper (PDF, DOI, arXiv link, or excerpt)
+2. The agent maps the macro skeleton in Stage 1
+3. Enter Stage 2 for mechanism deep-dive (`"进入阶段二"`)
+4. Enter Stage 3 for critique and reproduction roadmap (`"进入阶段三"`)
+
+## Who it's for
+
+- Master's and PhD students preparing to present or reproduce a paper
+- Researchers systematically surveying a field
+- Engineers extracting transferable methods from academic literature
+
+## Repository
+
+https://github.com/yarran-eng/literature-deep-reading

agents/yarran-eng__literature-deep-reading/metadata.json

@@ -0,0 +1,14 @@
+{
+  "name": "literature-deep-reading",
+  "author": "yarran-eng",
+  "description": "Rigorous 3-stage academic paper deep-reading & research mentor agent — macro skeleton, mechanism deep-dive, and reproduction roadmap with zero-hallucination epistemic rigour.",
+  "repository": "https://github.com/yarran-eng/literature-deep-reading",
+  "version": "1.0.0",
+  "category": "education",
+  "tags": ["academic", "research", "papers", "mentor", "llm", "feynman", "peer-review", "reproduction", "chinese", "english"],
+  "license": "MIT",
+  "model": "claude-sonnet-4-6",
+  "adapters": ["claude-code", "system-prompt"],
+  "icon": false,
+  "banner": false
+}