How builders put tools together to get work done.
gstack gives coding agents role-based workflows for planning, review, browser QA and release.
- Review stages. The workflow includes separate design and engineering reviews, with a verification command that runs the repository’s own checks and selected tests. Source ↗
- Browser QA. The reporting-only QA workflow collects replayable evidence and leaves product code and tests untouched. Source ↗
Scheduled · awaiting first completed check
GitHub Spec Kit turns a specification into plans, tasks and implementation through a supported coding agent.
- Core workflow. Spec-driven development starts with project principles and carries a feature through specification, planning, tasks, implementation and convergence. Source ↗
- Independent processes. Bug fixing and idea assessment are installable bundled extensions with their own entry points. Source ↗
Scheduled · awaiting first completed check
Foreman is a Vercel eve template that turns a GitHub, Linear or HTTP task into a reviewed draft pull request.
- Intake. A labeled GitHub issue, Linear delegation or HTTP ticket can start the pipeline. A person reviews and merges the resulting draft. Source ↗
- Workflow. Triage looks for duplicates and missing details, the analyst writes acceptance criteria, and the implementer runs repository checks in its own sandbox. Source ↗
Scheduled · awaiting first completed check
The Gemini cookbook search-grounding notebook streams a company report using the Gemini API and Google Search.
- Notebook runtime. The Colab example uses the Google Gen AI Python SDK and reads GEMINI_API_KEY from Colab Secrets. Source ↗
- Grounding. The notebook enables the Google Search tool, streams model output and renders Search Suggestions from grounding metadata. Source ↗
Scheduled · awaiting first completed check
GPT Researcher turns a research question into a cited report by planning subquestions, collecting sources and synthesizing findings.
- Pipeline. A planner proposes research questions, execution agents gather material, and a publisher combines the source-tracked findings. Source ↗
- Inputs. The workflow supports both web research and local documents. Source ↗
Scheduled · awaiting first completed check
BEBuilding effective agents examples
Anthropic’s orchestrator-workers notebook demonstrates model-driven task decomposition and execution by specialist prompts.
- Task decomposition. An orchestrator emits an analysis and XML subtasks, each with a type and a description selected at runtime. Source ↗
- Worker calls. The Python example calls each worker sequentially with the original task and worker-specific instructions. Source ↗
Scheduled · awaiting first completed check
Anthropic’s Computer Use Demo runs a Claude agent against a Linux desktop in Docker and displays the chat and desktop in a local web interface.
- Runtime. The container runs the agent loop alongside an X11 desktop, VNC and a Streamlit chat interface. Source ↗
- Session limit. The agent shares the desktop container and the demo supports one session at a time. Source ↗
Scheduled · awaiting first completed check
The Awesome LLM Apps customer-support example answers typed Streamlit questions from crawled documentation and returns text and synthesized speech.
- Indexing. Firecrawl collects documentation with a default five-page limit, and FastEmbed generates embeddings for Qdrant retrieval. Source ↗
- Query and playback. The Streamlit app submits text questions and offers playback and download of the generated speech. Source ↗
Scheduled · awaiting first completed check
Lesson 8 of Microsoft’s AI Agents for Beginners course compares ways to coordinate specialized agents.
- Coordination patterns. The lesson illustrates group chat, rule-based handoffs and collaborative filtering. Source ↗
- Operational visibility. It covers action logs, interaction graphs and metrics for latency, task volume and accuracy. Source ↗
Scheduled · awaiting first completed check
OCOpenClaw community use cases
The community’s Multi-Source Tech News Digest recipe describes how to collect, score and deliver a scheduled digest in OpenClaw.
- Inputs. The recipe combines RSS feeds, X accounts, GitHub releases and Brave Search queries. Source ↗
- Ranking. It describes deduplicating by title similarity and scoring for recency, source priority, corroboration and engagement. Source ↗
Scheduled · awaiting first completed check
SASelf-hosted AI starter kit
The Self-hosted AI starter kit packages n8n, Ollama, Qdrant and PostgreSQL for local workflow prototypes.
- Service layout. Docker Compose runs n8n with PostgreSQL backing storage, Ollama for inference and Qdrant for vector storage. Source ↗
- Bootstrap. The Compose setup imports a sample n8n workflow and pulls Llama 3.2 for the bundled Ollama service. Source ↗
Scheduled · awaiting first completed check
The OpenAI Cookbook portfolio notebook shows a lead agent calling specialist agents and combining their findings into a research memo.
- Agent control. The portfolio manager keeps control and calls macroeconomic, fundamental and quantitative specialists as tools. Source ↗
- Data tools. The example combines hosted Web Search and Code Interpreter with a local Yahoo Finance MCP server and FRED data functions. Source ↗
Scheduled · awaiting first completed check
Local Deep Researcher runs a language model locally and uses web search to produce a cited research summary.
- Research loop. The graph looks for gaps in its running summary and generates follow-up queries for a configurable number of rounds. Source ↗
- Local runtime. Ollama or LM Studio serves the model, and some steps expect structured JSON with fallback handling for incompatible output. Source ↗
Scheduled · awaiting first completed check
STORM explores a topic from multiple perspectives, then uses the resulting outline and references to draft a cited article.
- Research loop. Prewriting surveys related articles for perspectives and simulates exchanges with a source-grounded expert to generate follow-up questions. Source ↗
- Drafting. The writing stage uses the outline and collected references to produce an article with citations. The project describes the result as a draft that needs editing for publication. Source ↗
Scheduled · awaiting first completed check