mehdiforoozandeh
UserA scientific-method lab notebook your AI agent drives — research as a tree of questions and falsifiable hypotheses in plain markdown, with an LLM literature wiki, a task manager, and a local browser GUI cockpit.
Categories
Indexed Skills (19)
crux-cockpit
Launch and manage the crux cockpit — the read-only browser GUI over a crux research vault (`crux serve`). This skill runs it beginning to finish: locate the vault, start the server fresh on localhost, verify it actually serves, and hand the user one clickable URL — plus status / stop / restart, and a setup-or-demo path when no vault exists yet. Works wherever the agent runs: local machine, VS Code Remote-SSH (drives the port forward itself via `code --openExternal`), or a plain SSH terminal (hands the user the exact `ssh -L` tunnel command). Use when a crux user wants to see their vault in a browser. Triggers: "open the cockpit", "launch the crux GUI", "show me the tree in a browser", "serve the vault", "is the cockpit running", "stop the cockpit", "restart the cockpit", crux gui, crux cockpit.
crux-wiki
A literature wiki for a crux research vault — Andrej Karpathy's LLM-wiki pattern applied to a crux project. The PI curates immutable sources under `raw/`; you (the agent) compile them into a persistent, interlinked `wiki/` of background, prior methods, SOTA, baselines, datasets, and definitions, and draw on it to ask sharper questions and design better hypotheses. The crux engine owns the bookkeeping (source hashes, the generated `WIKI.md` index, structural lint via `crux validate`); you supply the reading and synthesis. One-way by design: literature → wiki → informs the tree; a project's own findings never enter the wiki. Use when a crux user wants a project-relevant literature base. Triggers: "ingest this paper", "add to the wiki", "what does the literature say", "compile a wiki", "lint the wiki", literature review, prior work / SOTA / baselines for a crux project, crux wiki, LLM wiki.
crux
An agentic research companion — a scientific-method lab notebook for navigating large research programs. Organize work as a tree of Questions (what we don't know) and falsifiable Hypotheses (testable leaves), each with pre-registered verifiables and findings; a deterministic engine rolls results up into per-question answers, trips a human review gate, and regenerates an Obsidian-graphable META + experiments registry. Use when the user wants to run a research program rigorously: open/track research questions, design experiments as problem-statement → hypothesis → verifiables → findings, synthesize results, and update the open questions. Triggers: crux, "open a research question", "lab notebook", "hypothesis/experiment tracking", "design an experiment", "what should we try next", "research notebook", scientific method, meta-questions, Obsidian research vault.
evolve-crux
Evolve crux itself — add a capability or fix a recurring flaw in the crux tool, end to end: ideate → build → validate → ship. Turns a feature idea or a "crux keeps doing X" annoyance into a signed-off PRD, a tests-first implementation, a hard validation gate (selftest green · stdlib-only · existing vaults still load · version/migration), and a pull request. Works for anyone contributing to crux, not just the maintainer; the maintainer self-merges and cuts releases. Use when working ON crux — adding to the engine/CLI/skill, picking up a ROADMAP epic (UI, wiki, marketing), or fixing an engine bug. Triggers: "evolve crux", "add a feature to crux", "crux keeps <doing X>", "improve crux", "contribute to crux", "send a crux PR", crux roadmap, wiki/GUI/autoresearch for crux.
brevity
Concise-and-clear response mode that maximizes the reader's understanding per word: maximize understanding, then minimize words — in that order. Keep full grammatical sentences (not clipped fragments), technical precision, and every fact; cut only filler, hedging, preamble, and recap. Use when there's a question to answer and the user wants it dense, clear, and scannable. Triggers: /brevity, "be concise", "keep it short", "no fluff", "tl;dr", "get to the point", "stop padding".
catchup
Brief refresher for resuming a long-running chat after time away. Reads back the current conversation and distills one scannable screen — what we're working on, where we are (including decisions already made), what's waiting on your call, and the next moves. Use when returning to a session after a break (a weekend, a holiday, a few days) and needing to re-orient before continuing. Triggers: /catchup, "where were we", "catch me up", "refresh me", "where are we", "remind me where we left off".
dialectic
Resolve a hard question by running a three-agent dialectic: a Thesis agent proposes, an Antithesis agent attacks/questions, a Synthesis agent sublates (preserves + transcends) into a new position that becomes the next round's thesis. Iterates until the Synthesis agent declares a clear answer, or escalates to the user at 15 rounds. Use when user wants an idea stress-tested to convergence, a genuine truth-seeking back-and-forth (not a debate to win), or a tough design/research/conceptual question worked from multiple independent minds. Triggers: /dialectic, "run a dialectic on…", "thesis/antithesis/synthesis this", "stress-test this to convergence".
era
Run ERA-style empirical-software search (Google ERA / Flat UCB Tree Search) to evolve whole programs toward a scalar objective. Use for software/algorithm DESIGN: an LLM writes and rewrites complete candidate programs, a sandbox scores each, and a flat PUCT bandit (FUTS) keeps a population and decides which to improve next — escaping local optima and returning a diverse portfolio of winners. Domain-agnostic. Triggers: ERA, FUTS, Flat UCB Tree Search, empirical software search, evolutionary program search, generate_fn/execute_fn.
hier
Hierarchical-outline response mode: answer as a nested-bullet toggle list (a tree), not prose. Use when user wants a response structured as a hierarchy — designing a project/system from scratch (goal → main components → subcomponents → …), breaking a problem down, mapping structure, or any answer where the shape is a tree and they want to scan it at a glance. Each node is a few words. Design/architecture hierarchies auto-render as a mermaid diagram instead. Triggers: /hier, "give me the outline", "as a hierarchy", "as a toggle list", "break this down", "structure this".
pingpong
Interactive, high-bandwidth collaboration mode: think deeply but answer in tight, one-idea-at-a-time turns that invite a reply. Use when user wants to brainstorm, design something from scratch, be grilled/Socratically questioned, pressure-test an idea, or do any open-ended collaborative work they want to keep engaging and iterative rather than receiving one large dump of text. Triggers: /pingpong, "let's brainstorm", "design this with me", "think out loud with me", "let's go back and forth".
quiz-me
Test whether the user's mental model still matches work that already exists — code the agent wrote, a PRD, a plan, a document tree — usually because they delegated it, skimmed it, or were away while it was built. Derives a tree from the artifact, then asks one multiple-choice question at a time: general framing first, then hard into the decision points, with distractors drawn from the alternatives actually rejected, or from the conventional choice this artifact departed from. Every wrong answer is triaged — did the agent drift from the user's intent (→ punch list), or is it a gap in the user's knowledge (→ explained)? Needs an artifact that already exists: for a subject nothing was built for use `tutor`, and for a plain re-orientation with no questions asked use `catchup`. Triggers: /quiz-me, "quiz me on what we built", "test me on this plan", "grill me on the code you wrote", "do I actually understand what we built", "am I still in sync with what you wrote", "I approved that without reading it".
resolve
Resolve a stuck trade-off — "we can't improve A without hurting B" — by finding a structural fix that dissolves the conflict instead of trading along it. Names the single element that must be two things at once, then looks for the escape across four separations (space, time, condition, whole-vs-part), a reparameterization (derive B from A), and a trim. Lists only the moves that genuinely apply and is honest when none do — says "genuine trade-off, here's how to pick" rather than dressing up "you have to choose" as an escape. Use on nascent/stuck design tensions, NOT mature problems where only tuning is left. Triggers: /resolve, "resolve this trade-off", "can't improve X without hurting Y", "stuck on a trade-off", "break out of this frontier", "the same thing must be both … and …".
tutor
Interactive multiple-choice teaching drill on a subject the user doesn't know yet. Requires a grounding source (the user names one, or it asks, or it searches out a credible one), shows the syllabus for approval, then climbs a general → specific knowledge tree one question at a time: right answers get a one-line why and the next question, wrong answers get a short correction and more questions around the gap. Fast back-and-forth, no walls of text, no memory between sessions. Use when the user wants to learn a subject by being asked rather than lectured — including "quiz me on X" where X is a subject rather than something we built. NOT for performing a task right now ("teach me how to fix this error"), and NOT open-ended brainstorming (use `pingpong`). Triggers: /tutor, "teach me X", "tutor me on X", "help me learn X", "crash course on X", "I don't know anything about X", "read this and teach it to me", "walk me up from the basics".
terse
Rapid-fire conversation mode: reply in 2–3 sentences max — usually one, sometimes two words — the way a colleague answers across a desk. A hard ceiling, not a target: the reply is sized to the question, and when the whole answer doesn't fit, the part that changes what the user does next lands now and the rest waits to be asked for. Use when firing off a queue of questions and wanting to keep the momentum of a real conversation. Triggers: /terse, "be super brief", "super short", "quick answers", "rapid fire", "short answers only", "keep it moving".
dispatch
Pick the model tier for every subagent before spawning it, so cheap work runs on cheap models: Opus for anything that writes, runs, implements, designs, or carries consequence; Sonnet for read-only work like research, reading, and summarizing; a Haiku swarm when fanning out more than five agents on a breadth sweep or brainstorm; and grok, on the separate Cursor meter, for token-heavy work whose answer you can check cheaply. Also checks the remaining Claude 5-hour and 7-day allowance and shifts checkable work to grok when either window runs low. TRIGGER — read BEFORE any subagent spawn, including agents you decided to spawn on your own initiative and `agent()` calls inside a Workflow. Do not skip it because the task "looks obvious" — the lane is a two-second check and the default is wrong half the time. Also fires on /dispatch, "spawn subagents", "fan out", "run these in parallel", "use a swarm of agents", "which model should this agent use", "route this to a smaller model".
orchestrate
Run a multi-step task end to end as a foreman: a planner writes a plan file with non-overlapping chunks, builder subagents build the chunks in their own git worktrees, and the orchestrator only briefs, reads return slips and diffs, merges, and re-briefs. The orchestrator never does big chunks of work itself, so its context stays fresh for the whole run. Does not stop until every acceptance criterion passes, or a chunk hits its retry cap. Routes every spawn through the dispatch skill. Use when the user wants a task done end to end by a group of subagents, not by the main loop. Triggers: /orchestrate, "orchestrate this", "run this end to end with subagents", "be the foreman", "delegate the whole thing and just review", "keep your context fresh".
dialectic-brainstorming
Brainstorm a consequential, no-ground-truth question with a panel of diverse minds, then resolve it dialectically. A casting step derives orthogonal thinking-axes for the question (each a quality to maximize) and casts one persona per axis; the personas propose and cross-critique concrete candidates while a neutral Synthesizer prunes, recombines across axes, and holds a frontier — returning a short list, not one winner. Use for naming (products, tools, projects), positioning, or any choice of consequence with no objectively correct answer where you want many genuinely different perspectives, not one. Triggers: /dialectic-brainstorming, "brainstorm this with a diverse panel", "what should we name…", "give me a shortlist from multiple perspectives".
latex-vscode
Set up, compile, preview, and troubleshoot LaTeX in VS Code with the LaTeX Workshop extension — an Overleaf-equivalent local/remote workflow (build-on-save, PDF preview, SyncTeX, latexmk, bibtex citations, LTeX grammar, Live Share). Use whenever working with .tex/.bib files, building a PDF, fixing a LaTeX/latexmk compile error, wiring citations/bibliography, or configuring the LaTeX Workshop / LTeX / Live Share setup. Triggers: latex, .tex, latexmk, pdflatex, lualatex, bibtex, \cite, build the pdf, compile my paper, SyncTeX, Overleaf, LaTeX Workshop. NOT for writing prose (use scientific-writing) or generating BibTeX entries (use citation-management).
mansplain
Re-situate something you (the agent) just said when it drifted into jargon or internal framing the user never grounded — NOT by dumbing it down (ELI5), but by filling in the missing context: a brief backdrop, how it fits together, and where it lands in *this* conversation, always with 1–2 simple concrete examples. Governed by one rule — the certainty test: say and explain only what you are not sure the user already knows. Use when the user asks to have something put in perspective, re-anchored, or explained in the conversation's own terms. Triggers: /mansplain, /mansplain <term>, "you lost me", "put that in perspective", "situate that", "wait — what actually is X and why does it matter here".
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.