Weekly Trending

Brainhuggers Bureau
← PrevKW30 · 20.07–26.07.2026Next →
GitHub Trending
01
block buzz
+2.5k this wk · 6.6k stars
block/buzz
A self-hostable human-and-agent workspace built around shared rooms, signed events, workflows, git events, search, and an agent-first CLI.
02
ayghri i-have-adhd
+1.8k this wk · 6.7k stars
ayghri/i-have-adhd
A portable agent skill that pushes AI assistants toward brief, direct, low-friction responses.
03
KnockOutEZ wigolo
+605 this wk · 1.8k stars
wigolo: local-first web research for AI coding agents
wigolo is a local-first MCP layer for web search, fetch, crawl, and research that gives coding agents web access without relying on cloud API keys.
04
LangChain open_deep_research
+14 this wk · 7.6k stars
Deep research becomes an inspectable workflow graph
LangChain's open_deep_research is an open-source LangGraph deep research agent with explicit stages for clarification, research brief writing, supervisor delegation, researcher loops, compression, and final report generation. The repository exposes provider, search, MCP, state, and evaluation surfaces rather than hiding deep research inside one prompt.
05
theopenco llmgateway
+6 this wk · 1.5k stars
theopenco/llmgateway
llmgateway is an open-source gateway for routing, managing, and analyzing LLM requests across providers.
06
makenotion notion-mcp-server
+4 this wk · 4.6k stars
makenotion/notion-mcp-server
Notion's official MCP server exposes Notion workspace data and actions to agents through a governed tool interface.
HF Papers
01
Rethinking the Evaluation of Harness Evolution for Agents
Rethinking the Evaluation of Harness Evolution for Agents
The paper argues that agent harness evolution can overstate progress when the same task surface is used for both search feedback and final evaluation, and calls for held-out tasks and matched test-time scaling baselines.
02
RESOURCE2SKILL
RESOURCE2SKILL: Distilling Executable Agent Skills from Human-Created Multimodal Resources
The paper proposes a system for turning human-created tutorials, repositories, articles, and multimodal artifacts into executable skills that agents can retrieve, compose, and improve.
03
Behavioral Privacy Leakage in Agentic Negotiation
Behavioral Privacy Leakage in Agentic Negotiation
The paper formalizes inference attacks where negotiation agents reveal private constraints through concession timing, rhythm, convergence behavior, and randomized policy patterns even without stating the private values directly.
04
Self-State Attacks
Self-State Attacks on Self-Hosted AI Agents: How Far Can OS Defenses Go?
The paper studies attacks against self-hosted AI agents by corrupting their own memory, configuration, and state files through allowed operating-system actions.
05
Beyond Relevance-Centric Retrieval
Beyond Relevance-Centric Retrieval
A retrieval paper arguing that document sets should be selected and ranked by coverage, contradiction, complementarity, and redundancy, not only individual relevance scores.
06
AREX
AREX: Towards a Recursively Self-Improving Agent for Deep Research
AREX is a deep-research agent architecture that recursively audits provisional answers against constraints, keeps verified evidence, and launches follow-up research for unresolved claims.