Universal ARA Compiler. Converts ANY research input — PDF papers, GitHub repositories, experiment logs, code directories, raw notes, or combinations thereof — into a complete Agent-Native Research Artifact (ARA): a structured, machine-executable knowledge package with a cognitive layer (claims, concepts, methods), an artifact layer (code/configs/data as the work warrants), an exploration graph (research DAG), and grounded evidence. Works across any research field — not only model-training research. TRIGGERS: compile, create ARA, generate artifact, convert paper, build artifact, compile paper, ARA from PDF, ARA from repo, ARA from code, structure research, extract knowledge, extract figure data, digitize plot, read chart, figure to data
ARA World Model — read-only reasoning engine over ONE Agent-Native Research Artifact (ARA), run LOCALLY with the coding agent itself as the LLM (no SDK, no API key). Given an ARA directory and a free-text query, it answers any question about the ARA — a forward "what if I change X", but equally why-did-this-work, what-should-I-try, is-this-sound, how-do-these-compare, or anything else — by retrieving precedent from the ARA's native files (references/RETRIEVE.md) and answering as the Predictor (references/PREDICT.md): a bold, grounded, falsifiable Answer shaped to what the question actually calls for. TRIGGERS: ask the world model, wm predict, predict with the world model, what if I change X, forecast the loss curve, will this help, why did this work, what should I try next, is this claim sound, compare these, retrieve precedent, what precedent surfaces.
End-of-turn research process recorder with progressive crystallization. Invoked at the END of EVERY turn, after the user's current request has been fully addressed and before yielding control back to the user. Reviews what happened in the turn, extracts research-significant events, and writes them into the ara/ artifact through a three-stage pipeline: Context Harvester → Event Router → Maturity Tracker. Trace events (decisions, experiments, dead ends, pivots) are recorded immediately as journey facts. Knowledge events (claims, heuristics, concepts, constraints) are staged first and crystallize into typed layers ONLY when closure signals appear — topic abandonment, verbal affirmation, empirical resolution, or artifact commitment. NEVER mid-turn. All entries carry provenance tags (user / ai-suggested / ai-executed / user-revised).
Research Visualizer. Renders an existing Agent-Native Research Artifact (ARA) into ONE self-contained, interactive HTML file showing the AI scientist's step-by-step research process: a clickable process map of the exploration tree (branches and dead ends included) on the left, and a per-step drill-down on the right — what the step did (its narrative written in plain language a person can follow), why (the linked claim), the real result (verbatim grounded numbers + inline figures + tables), and the code/artifact pointer. Read-only consumer of the artifact — it never changes how research is done. When the ARA carries them, it also surfaces (each optional, only when present) the related-work dependency graph, the problem framing, a concepts glossary with in-text term popovers, and the solution recipes — reached from header disclosures without leaving the process map. Accepts either an existing ARA or raw research input (a paper, repo, run logs, or notes); when the input is not yet an ARA it is compiled into one
ARA Submitter / Publisher. Takes a research directory, makes sure it is a valid Agent-Native Research Artifact (ARA) — compiling or updating it with the `compiler` skill when it is not — guarantees it carries an interactive visualization (running `research-visualizer` to produce `trajectory.html` when missing), then publishes it as a public GitHub repository on the authenticated user's own account and links it into the ARA Hub website so others can browse and replay it. GitHub is the data layer; the Hub fetches from it. TRIGGERS: submit, submit ara, publish ara, upload ara, share ara, push ara to github, add to ara hub, submit-ara, publish artifact, make my ara public
ARA Seal Level 2: Semantic Epistemic Review. Acts as an objective research reviewer for Agent-Native Research Artifacts. Assumes Level 1 structural validation has already passed. Evaluates six dimensions of epistemic quality through semantic reasoning over the ARA's content. Produces a scored review with per-dimension strengths/weaknesses/suggestions, severity-ranked findings, and an overall recommendation (Strong Accept to Reject). TRIGGERS: level2, seal level 2, verify level 2, epistemic audit, review ara, audit claims