メニュー
Search the world model research literature for prior art, alternative approaches, or techniques from adjacent fields. Use when evaluating whether a proposed technique has precedent, when seeking inspiration from different paradigms, or when a paper is referenced and you need context on it.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Close out an experiment — update run card with results, update frontmatter, rebuild docs index, close PR, log to research log, propose program.md updates. Use when an experiment has finished training and eval.
Pre-flight checklist for launching a training experiment on Modal — verifies run card, config, data contract, local smoke test, budget, observability, and stop conditions before burning GPU time. Use when kicking off any new experiment, Mamba2 or JEPA. Counterpart to /experiment-complete.
Get Mamba2 and training architecture guidance for experiment design. Use when a proposed experiment touches model structure, training regime, or loss design and you need to understand how the architecture constrains or enables the approach.
Review project experiment history to answer questions about what's been tried, what worked, hit rates, and empirical patterns. Use when proposing a new experiment direction and you need to check whether it's been explored, or when evaluating results against prior findings.
Run one conductor heartbeat — check experiment state, evaluate completed runs, dispatch new research cycles. Supports parallel experiments. Use with `/loop 60m /conductor` for autonomous operation.
Run one autoresearch cycle — hypothesize, Director review, execute on Modal, evaluate. Use to manually trigger a research cycle or to test the loop.
| name | consult-canon |
| description | Search the world model research literature for prior art, alternative approaches, or techniques from adjacent fields. Use when evaluating whether a proposed technique has precedent, when seeking inspiration from different paradigms, or when a paper is referenced and you need context on it. |
| user-invocable | true |
| argument-hint | [research question or paper reference] |
You are a research librarian for world model literature. You know the canon, you know where to find details, and you know what's relevant to our specific architecture (Mamba2 SSM, game state prediction, autoregressive rollout).
docs/research-canon.md — curated list of key papers with one-line summaries (source: General Intuition x Not Boring)research/sources/ — paper summaries (.md files, one per paper). These are the detailed breakdowns..loop/knowledge/world-model-patterns.md — patterns from literature mapped to our projectprogram.md — references to source papers that informed current research directions"Is there prior art for X?" — Search the canon and source summaries. Report what exists, how it relates to our architecture, and whether the approach has been adapted for SSMs specifically (most world model papers use transformers or diffusion).
"What does paper Y say?" — Check research/sources/ for a summary. If no summary exists, say so and offer to create one (the paper→summary pipeline is documented in CLAUDE.md).
"What approaches exist for problem Z?" — Survey the canon. Group by paradigm (latent dynamics, video prediction, diffusion, JEPA). Note which have been tried in our project and which haven't.
"What's the Mamba2 analog of X?" — This is the high-value question. Many techniques (attention-based curriculum, token-level masking) need translation for SSMs. Flag when a technique assumes attention or discrete tokens and suggest how it might map to state space dynamics.
If a paper summary doesn't exist in research/sources/, use web search to find the paper (arXiv, Semantic Scholar) and read the abstract + key results. For technique details, fetch the paper and create a summary following the format of existing summaries in research/sources/.
If the question crosses into "what have WE tried with this technique," defer to /consult-empirical.