Open-Source AI Repos & MCP Servers

An open-source development environment for Claude Code — often called 'oh-my-zsh for Claude Code'. One 5-minute install wires up 11 specialized agents, 33 slash commands, 24 skill workflows, 15 automation hooks, and a 6-layer security-hook system, turning a bare CLI into a pre-wired plan-to-deploy pipeline.

A Claude Code skill suite that stops the AI from faking citations. 22,700 stars. Latest release v3.9.2 ten days ago. CC BY-NC 4.0 — free for academic use, separate license required for commercial. Imbad0202/academic-research-skills is the Claude Code skill suite for the full academic research pipeline. Python (97.4%). Latest release v3.9.2 published May 18, 2026. Trending on GitHub the week of May 21, 2026. CC BY-NC 4.0 — commercial users need a separate license. The Citation-Hallucination Problem It Solves Three university scholars have been sanctioned in 2026 for submitting papers with LLM-fabricated citations to papers that do not exist. The failure mode is well-documented — Zhao et al. 2026 and Lu et al. 2026 are the peer-reviewed studies the maintainer cites in the repo. The pattern: an LLM is asked to find supporting literature, returns plausible-looking citations, and the researcher does not verify them before they ship. By the time the reviewer catches it, the paper is rejected and the scholar's name is in a retraction database. Academic Research Skills attacks the failure mode directly. The reference-hallucination detection stack verifies every cited paper against three independent academic indexes — Semantic Scholar, OpenAlex, Crossref — before the citation lands in the draft. Temporal verification catches LLM-fabricated publication dates. Claim-faithfulness auditing checks that the cited paper actually supports the claim being made. The agent stops being the source of the failure and starts being the layer that defends against it. The Four Coordinated Skills Deep Research. Systematic literature synthesis with systematic-review support. Socratic-dialogue mode for hypothesis development. Cross-index verification across Semantic Scholar, OpenAlex, and Crossref. The output is a literature review the human can defend in a viva. Academic Paper. Multi-agent writing with style calibration that learns the author's voice from prior work. Writing-quality assessment against a published rubric. Format conversion across APA 7.0, Chicago, IEEE, MLA, and Vancouver in one command without breaking citation integrity. Submitting to a new journal becomes a 30-second command instead of a 2-hour reformatting slog. Academic Paper Reviewer. Multi-perspective evaluation with 0-100 quality rubrics. Devil's Advocate adversarial mode that surfaces the criticisms a real peer reviewer will raise. Run your draft through it before you submit, fix what it flags, and your reviewer-response cycle gets shorter. Academic Pipeline. Ten-stage orchestrator that chains the other three skills with integrity gates, claim verification, and a Material Passport that tracks every cited source through the workflow. The orchestrator is what makes the suite a full lit-review-to-submission tool rather than four disconnected scripts. The 'AI Is Your Copilot, Not The Pilot' Philosophy The maintainer's design philosophy is explicit and anti-autonomy. The skills are built to assist a researcher who is driving the work, not to replace the researcher. Human-in-the-loop is a feature, not a fallback. Every citation is auditable. Every claim is verifiable. Every output is reviewable. The pattern is the opposite of the agentic-replacement narrative that has dominated AI marketing for 18 months — the human writes the paper, the skills make the agent's contributions trustworthy enough to ship. This is the right design for academic work. Peer review is unforgiving. A single fabricated citation is enough to torpedo a career. The 'agent does it autonomously, you check at the end' workflow has already produced three sanctioned scholars this year. The right architecture is constrained-agent, audited-output, human-driven — and that is what the suite ships. The Money Math For A PhD Candidate The audience is paying $30,000-60,000/year in tuition. The dissertation is the multi-year deliverable that justifies the entire program. An LLM-fabricated citation flagged in defense or peer review is a career-defining event. Against that downside, a free (for academic use) skill suite that defends against the exact failure mode is the install of the decade. For non-students: the suite also replaces $500-1,500 of paid literature-review software (Covidence, Rayyan, EndNote citation management) and the hours of manual citation-checking that catches LLM fabrications. The integrated workflow is faster than running those tools separately, and the citation-verification quality is at least as good for the canonical-indexed coverage. The Commercial Use Caveat The CC BY-NC 4.0 license is the constraint to be aware of. Free for non-commercial academic use, including PhD students, postdocs, university researchers, unfunded individual research, and most grant-funded academic work. Commercial use — including in-house industry research labs that ship work into commercial products, paid consulting deliverables, or for-profit white papers — requires a separate license from the author. Read the LICENSE file before deploying inside a company. The free academic tier is generous; the commercial restriction is real. What To Do With It This Week If you are a PhD student or postdoc, install it. The install is two commands: /plugin marketplace add Imbad0202/academic-research-skills then /plugin install academic-research-skills. Requires Claude Code v3.7.0+. If you are a technical writer in an industry research lab, read the license first — commercial use needs a separate agreement. If you have ever shipped an LLM-assisted paragraph with a citation you did not personally verify, this is the install that closes that hole. For broader context on the AI-tools-getting-trustworthy pattern, see CodeGraph applying the same human-in-the-loop architecture to coding workflows and Pi Coding Agent as the model-agnostic harness that gives the human authority. The pattern across the stack is consistent — constrain what the AI is allowed to do, hand the human the steering wheel. Our breakdown this week of why the AI-replacement narrative is a sales pitch walks through the data on why this architecture wins.

A Claude Code skill that hits 96.7% task success with the skill enabled and 6.7% without. Same benchmark. 14.4x improvement. 22,100 stars. The latest release shipped yesterday. OthmanAdi/planning-with-files is the open-source implementation of the workflow pattern that powered Manus's reported $2 billion acquisition — teach the AI agent to treat the filesystem as memory. The skill is MIT-licensed, cross-compatible with Claude Code, Cursor, Codex, Gemini CLI, and 40+ agents on the agentskills.io open spec. Install with one npx command. The Three Markdown Files That Change The Agent When the skill is active, the agent writes and maintains three files in your repo. task_plan.md is the roadmap — phases, decisions, progress markers, the plan the agent committed to. findings.md is the research log — discovered constraints, dead ends, third-party gotchas, what the agent learned the hard way. progress.md is the completion log — what got done, when, with what result. The skill instructs the agent to reread all three before every decision, log errors back to findings.md for future reference, and verify task completion against task_plan.md. The agent's memory stops being a black box inside its context window and becomes an auditable record in your repo. Why The 14.4x Number Lands The 96.7% versus 6.7% benchmark is real and reproducible — the methodology is published in the repo. Without the skill, the same agent loses the plot halfway through complex tasks, forgets earlier decisions, and rebuilds wrong files from scratch. With the skill, every decision is anchored to a re-read of the plan and the findings. The agent stops drifting. For agencies billing $200/hour that re-prompt failed Claude tasks five times in a session, the math compounds fast. A 14.4x success rate on the same task set means roughly 14x less re-prompt waste — direct savings on Claude or Cursor seat usage. The skill pays for itself in four billable hours and then keeps paying. The Killer Feature: /clear Recovery The most-loved feature in the issues thread is the recovery layer. When your Claude Code context window fills and you /clear the session, the agent normally loses everything. With planning-with-files, the markdown files in your repo serve as the recovery scaffold. The next prompt rereads task_plan.md, findings.md, and progress.md, and the agent resumes where it left off — with all the prior decisions and discovered constraints intact. This solves the single most frustrating failure mode of long-horizon agent work. You no longer lose three hours of agent context because you hit a token limit on a Tuesday afternoon. Installation Takes Twelve Seconds Run npx skills add OthmanAdi/planning-with-files --skill planning-with-files -g and the global skill is live. The latest release v2.43.0 shipped on May 26, 2026 — one day before this listing — so the maintainer cadence is genuinely active. MIT license, no enterprise license needed for commercial use. What To Do With It This Week If you run any multi-step coding work through Claude Code or Cursor longer than a single afternoon, install it. If your team is paying Manus money and would prefer to keep that budget in-house, install it. If you have ever screamed at your AI agent for forgetting what it agreed to do twenty minutes earlier, install it. For broader context on the AI-agents-getting-real-memory trend, see related coverage of Anthropic's Project Glasswing grounding security agents in real partner code, and the Claude product line driving the agentic playbook the skill is built on top of.