autoresearch

What it does

Autoresearch orchestrates end-to-end autonomous AI research projects using a two-loop architecture. The inner loop runs rapid experiment iterations with clear optimization targets. The outer loop synthesizes results, identifies patterns, and steers research direction. Routes to domain-specific skills for execution, supports continuous agent operation via Claude Code /loop and OpenClaw heartbeat, and produces research presentations and papers. Use when starting a research project, running autonomous experiments, or managing a multi-hypothesis research effort.

How it works

The skill instructs the agent to operate as a research project manager rather than a domain expert. It mandates autonomous, ongoing progress without user confirmation, with frequent human-visible progress via HTML/PDF presentations. It defines a two-loop architecture: INNER LOOP for fast, hypothesis-driven experiments and OUTER LOOP for synthesis and direction changes. It provides a structured workspace layout and a process to route execution to domain-specific skills when needed. It emphasizes maintaining state files (research-state.yaml, research-log.md, findings.md) and a persistent heartbeat mechanism (via Claude Code /loop and OpenClaw) to keep operations continuous. It requires setting up an agent continuity loop at startup and provides steps for initializing and updating literature, hypotheses, evaluations, and results across a multi-hypothesis project.

When to use it

Use when starting a research project, running autonomous experiments, or managing a multi-hypothesis research effort. It guides whether to bootstrap from vague ideas, a clear research question, an existing plan, or resuming from an existing state (research-state.yaml).

What it can touch

Routes to domain-specific skills for execution when domain tasks are needed, including data processing, model training, evaluation, and infrastructure as specified by its internal routing guidance. It also references a set of repository locations and templates for project folders and files (e.g., research-state.yaml, research-log.md, findings.md, literature/). It supports continuous operation via Claude Code /loop and OpenClaw heartbeat to maintain ongoing progress.

Caveats

The skill is designed for autonomous operation and requires trust in its loop-driven progression. It emphasizes frequent progress reporting and human review via generated presentations, but does not guarantee final results. It specifies a mandatory agent continuity setup and continuous operation mechanisms to avoid stalling. License information in the frontmatter lists MIT for this skill.