MCP servers and skills: search, filter by where they run, and compare.
LLM code-agent framework — Thought→Code→Observation loop, process-isolated sandbox, MBPP + SWE-bench tracks, MCP tools, multi-model OpenRouter benchmarks. 42 School project.
PythonMIT★ 2Few starsNo key
Local-first MCP server that tells coding agents which files to read, and how sure it is. 97% Hit@10 on Java (SWE-bench Multilingual), 89% on Python (SWE-bench Lite), zero LLM cost. mcp-brain calibrate
PythonMIT★ 5No key
PythonMIT★ 2Few starsNo key
Local-first, auditable Python code agent. Ships its own 30-task hidden-test benchmark plus SWE-bench Verified scored by the official harness -- every number reproducible, negative results included.
PythonMIT★ 154No key
Fast, cheap judgment for AI coding agents: semantic search, focused reads and list picking in ~2s. CLI + MCP server on TypeSafe Jev. Benchmarked on SWE-bench.
PythonMIT★ 7No key
Playwright for coding agents. Test that your skills, MCP servers, and CLIs actually work when an agent uses them — sandboxed YAML suites, A/B experiments, CI gates.
PythonApache-2.0★ 154No key
Zero-LLM structural code retrieval for AI coding agents — tree-sitter index, BM25+dense+structural fusion, served over MCP.
PythonMIT★ 7No key
Verifiable AI coding agent harness / 可验证的 AI 编码代理：evidence hash-chain · hollow-answer gate · manual verification checklist · 6 engines + MCP · multi-model · Windows/CJK · listed in awesome-opensource-
PythonMIT★ 54No key
Architectural Moat: JIT-JEV Context OS — Epistemic runtime & JEV System 1 context gate for AI agents (L0 SQLite WAL <3ms, Epistemic Invariants I1–I10, CERN Zenodo DOI: 10.5281/zenodo.22649542)
PythonMIT★ 8No key
Placebo-controlled benchmark of the most-starred coding agent skills
SkillMIT★ 19No key
PythonMIT★ 7No key
Git-native context layer for AI coding agents. Your team's real engineering decisions — patterns, pitfalls, conventions — live as reviewed markdown files in your repo; agents consult them before writi
PythonMIT★ 27No key
Context engine for coding agents. Indexes a repo into a multi-layer knowledge graph and serves precise architectural context over MCP. 78.2% on SWE-bench Verified at $0.22/instance. MIT.
PythonMIT★ 2Few starsNo key
The acceptance layer for bug fixes: a failing test sealed for every bug, then used to grade every PR that claims to fix it — human or AI agent.
PythonMIT★ 5No key
Next-Generation Distributed Multi-Agent Runtime & Swarm Substrate uniting Super-Memory, Cyclical Swarms, 24/7 Durable Daemons, Multimodal Streaming, VLA Safety Governors, and Benchmark Evaluators.
PythonApache-2.0Few starsNo key
Transparent feedback compression middleware for LLM coding agents
PythonMIT★ 1Few starsNo key
Robot Framework library for testing MCP servers, Agent Skills, Hooks, SubAgents, and coding-agent CLIs — provider-agnostic via LiteLLM, BFCL-grade tool-call matching, statistical N≥10 by default.
PythonApache-2.0★ 1Few starsNo key
Agent triage, measured: does this edit need a human, or can the AI handle it? We measure whether the graph under the agent can reach the tests that guard a change (0.42 pooled, 7 repos) and ship the f
PythonMIT★ 2Few starsNo key
Community-driven behavioral reliability benchmark for LLMs. 231 probes across 19 modules, deterministic scoring, perplexity correlation, layer sensitivity mapping, quant method capture, hardware-strat
PythonMIT★ 9No key
Give your coding agent the right code, not more code. Local context engine for Claude Code, Cursor & Codex: code graph + hybrid search + embeddings + git history, packed into a token budget. Held-out
PythonMITFew starsNo key
One MCP tool that returns the files a coding agent should read.
PythonMIT★ 1Few starsNo key
Test-Driven AI Development: GraphRAG-powered test impact analysis that reduces AI code regressions by 72%
SkillMIT★ 2Few starsNo key
Score Claude Code skills with deterministic checks, safety gates, eval fixtures, and ship/revise/reject scorecards.
SkillMIT★ 1Few starsNo key
Who stands behind it
