kimtth/add-new-entry-from-temp-md

Workflow: Adding New Entries from temp.md

temp.md is the raw input — an unformatted checklist of URLs and short notes. The goal is to produce temp_entries.md as a properly formatted staging file ready to paste into the target section files.

Steps in order:

1. Classify each URL → determine which section file (azure.md, applications.md, models_research.md, best_practices.md, tools_extra.md) and which current section heading it belongs to.
2. Fetch descriptions — use code/fetch_github_description.py for GitHub repos. For arXiv papers and blog/web links, use fetch_webpage to extract a one-sentence description.
3. Fetch creation dates — use code/update_github_dates.py for GitHub repos. For arXiv, derive the date from the ID prefix (e.g., 2602.xxxxx → Feb 2026). For blog posts, read from the page.
4. Add star badges — use code/add_github_stars.py for all GitHub links.
5. Apply legend symbols — see the Legend Format section below. azure.md should not use emoji markers.
6. Shorten descriptions — keep each description to ≤15 words. One punchy sentence. Do not repeat the link name.

Legend Format

azure.md — dash-bullet, no emojis

- [Name](url) - Description. (Mon YYYY) ![stars](...)

• Do not use emoji markers in azure.md (no link-prefix emojis and no description-prefix emojis).
• Date is in (Mon YYYY) parentheses format with no brackets.
• Star badge goes at the end of the line, after the date.

Examples:

markdown
- [Azure ML Prompt Flow](https://learn.microsoft.com/...) - Visual designer for prompt orchestration and evaluation. (Jun 2023)
- [APIM-Sample](https://github.com/Azure-Samples/APIM-Sample) - Single APIM endpoint for multiple models. (Jan 2026) ![**github stars**](...)

applications.md, models_research.md, best_practices.md — numbered list (or dash), symbol appended to link text

1. [Name](url): Description. [Mon YYYY] ![stars](...)

or (for entries that use dash bullets in that section):

- [Name](url): Description. [Mon YYYY]

• The legend symbol is appended inside the link text, immediately after the name (no space before the symbol).
• Date is in [Mon YYYY] square-bracket format.
• Star badge goes at the end of the line, after the date.
• Use numbered list (1.) when the surrounding section uses numbered lists; dash (-) when not.

Examples:

markdown
1. [Auto-Claude](https://github.com/AndyMik90/Auto-Claude): Autonomous multi-session AI coding. [Dec 2025] ![**github stars**](...)
1. [Towards AI Search Paradigm📑](https://arxiv.org/abs/2506.17188): Modular 4-agent system using DAGs for retrieval-intensive search. [Jun 2025]
- [Claude Code Security](https://www.anthropic.com/news/claude-code-security): Claude Code on the web for scanning codebases. [Feb 2026]

Legend Symbols

| Symbol | Meaning | |--------|---------| | | Blog post / documentation / web page | | 📑 | Academic paper (arXiv) | | 📺 | Video content | | 🤗 | Hugging Face resource |

Section Reference

Use exact heading names when labeling entries in temp_entries.md. Format: ## <filename> - <Section Name>:.

azure.md

• Azure OpenAI & Foundry Overview
• Orchestration Frameworks
• Prompt Engineering & Tooling
• Agent Frameworks
• Model Training & Inference
• Safety, Security & LLMOps
• Data Processing & Memory
• Dev Tools, MCP & Extensions
• Copilot Product Catalog
• Microsoft Foundry & AI Services
• Azure AI Search
• Agent Development
• Microsoft 365 Agent Development
• Learning Resources & Workshops
• Microsoft Research
• Sample Applications
• Solution Accelerators
• Code Samples & Workshops
• Architecture Patterns & Use Cases

applications.md

• RAG (Retrieval-Augmented Generation)
• GraphRAG
• RAG Application
• Vector Database & Embedding
• Top Agent Frameworks
• Additional Agent Framework
• Cache
• Data & Analytics Agents
• Data Processing & OCR
• Desktop AI assistant
• Memory
• Model Gateway
• Model Serving & Local Runtimes
• Observability & LLMOps
• SDKs, Integration & ML Libraries
• Training & Fine-tuning
• UI & No-Code Tool
• A2A
• Computer use
• Model Context Protocol (MCP)
• Coding
• Deep Research
• Domain-Specific Agents
• Skill
• Harness

> Tip: Do not add hand-curated entries to generated index sections such as Popular LLM Applications (GitHub Stars >= 1000); update the generator skill instead.

models_research.md

• Large Language Model Landscape
• Large Language Model Comparison
• Taxonomy of Natural Language Processing
• LLM Evolution and Model Timelines
• NLP Taxonomy and Research Fields
• Large Language Model Collection
• Architecture Comparisons
• Foundation Model Providers
• Domain-Specific and Specialized LLMs
• Multimodal Models
• Prompt Engineering and Visual Prompts
• Prompt Engineering
• Adversarial Prompting
• Prompt Tuner and Optimizer
• Prompt Guides and Prompt Libraries
• Visual Prompting and Visual Grounding
• Large Language Model Training and Optimization
• Pre-training and Data Preparation
• Architecture and Inference Patterns
• Architecture Variants, Attention, and Sparse Computation
• Context and Long-Context Limits
• Tokenization and LLM Numbers
• Capabilities and Evaluation
• Reasoning
• Post-training and Fine-Tuning
• Model Merging and Continual Adaptation
• Parameter-Efficient Fine-Tuning
• LoRA: Low-Rank Adaptation
• Alignment and Preference Optimization
• Quantization Techniques
• Pruning and Sparsification
• Knowledge Distillation
• Memory Optimization
• AI Adoption, Impact, and Society
• AGI, Society, and Long-Term Impact
• Trust, Safety, and Security
• Business Adoption and Use Cases
• Model Roadmaps and Products
• OpenAI Products
• Anthropic AI Products
• Google AI Products
• Survey on Large Language Models
• Additional Topics: A Survey of LLMs
• LLM Research (Ranked by cite count >=150)
• Learning Resources, Implementations, and Regional Materials
• Build LLMs from Scratch
• Japanese and Korean-Language Materials
• General Learning and Supplementary Materials

best_practices.md

• The Problem with RAG
• RAG Solution Design
• Agent Research → ### **Agent Research**
• RAG Research → ### **RAG Research**
• Agent Design Patterns → ### **Agent Design Patterns**
• Tool Use
• Tool Use: LLM to Master APIs
• Proposals & Glossary

tools_extra.md

• LLM for Robotics
• Awesome demo
• Datasets for LLM Training
• Evaluating Large Language Models
• LLM Evalution Benchmarks
• Evaluation Metrics
• LLMOps: Large Language Model Operations

Code Tools Reference

All tools are in code/. Run with python code/<script>.py.

| Script | Purpose | |--------|---------| | fetch_github_description.py | Fetch GitHub repo descriptions; appends after the link colon. Skips lines that already have a description. | | update_github_dates.py | Fetch GitHub repo creation date; appends [Mon YYYY] or (Mon YYYY). Skips lines already dated. | | add_github_stars.py | Append star badge to lines with GitHub links. Skips duplicates. | | fetch_popular_papers.py | Query Semantic Scholar for review-only RAG/agent paper candidates; not part of normal entry insertion. | | fetch_llm_papers.py | Generate or refresh the separate LLM-landscape paper pool; use fetch-llm-papers rather than hand-editing its output. | | update_citation_counts.py | Update citation counts for ranked paper sections via Semantic Scholar. | | check_unused_files.py | Scan markdown for file refs; move unreferenced files to files/_bak/. |

For arXiv papers and blog posts, fetch_github_description.py does not apply. Use fetch_webpage (agent tool) to retrieve a description from the URL.

Common CLI pattern:

powershell
python code/fetch_github_description.py --input temp.md --output temp_with_desc.md
python code/update_github_dates.py --input temp_with_desc.md --in-place
python code/add_github_stars.py --input temp_with_desc.md --in-place

Common Pitfalls (Lessons Learned)

1. Wrong legend placement: In azure.md, do not use emoji markers at all. In all other files, the symbol is appended to the link name inside [Name]. Never mix these two formats.

1. Wrong section names: Section labels in temp_entries.md must match the actual heading text in the target file exactly. Check the file before assigning. Do not invent new section names.

1. Missing descriptions for non-GitHub links: fetch_github_description.py only works for github.com URLs. For arXiv, blog, and product pages, you must fetch the page and write a description manually.

1. Verbose descriptions: Keep descriptions to ≤15 words. Do not repeat the name. No trailing "for use with", "that helps you", or similar filler.

1. Date format mismatch: azure.md uses (Mon YYYY) parentheses. All other section files use [Mon YYYY] square brackets.

1. Emoji preservation: Ensure edits preserve existing legend symbols such as 📑, 📺, and 🤗.

1. Star badges on non-GitHub links: Only add star badges to github.com links. Blog posts, arXiv papers, and product pages must not have a star badge.