autoresearch

autoresearch is a command for Claude Code from drivelineresearch/autoresearch-claude-code. It costs 14 tokens per session (544 once invoked), scanned A, original, MIT.

A command for starting, resuming, pausing, or inspecting an autonomous loop that runs software experiments and tracks their results.

In plain words
What is it for?
Use it with arguments such as off or status to pause the loop or inspect its dashboard and experiment log.
Why use it?
It makes it possible to stop the loop safely or review its progress without accidentally launching more experiments. Status information can include runs, outcomes, the best result, and whether the loop is active.

Command for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: reads .claude/ paths.

Runs only inside its plugin — its command needs a path that Claude Code sets for a plugin’s own hooks and for nothing else. Install the plugin, not this.

Part of the autoresearch plugin — 1 skill, 1 command, 4 hooks shipped together

Good fit Use it with arguments such as off or status to pause the loop or inspect its dashboard and experiment log.

Compare 6 commands from other repositories ↓
Install

Getting it into your agent

This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.

Claude Code
/plugin marketplace add drivelineresearch/autoresearch-claude-code
Claude Code
/plugin install autoresearch

Made for: Claude Code.

Or install autoresearch, the plugin that ships this one along with the rest of its 1 skill, 1 command, 4 hooks.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for autoresearch

README.md
[![agentmods](https://agentmods.dev/badge/commands/drivelineresearch/autoresearch-claude-code/autoresearch/github.svg)](https://agentmods.dev/commands/drivelineresearch/autoresearch-claude-code/autoresearch)
Your own site
<a href="https://agentmods.dev/commands/drivelineresearch/autoresearch-claude-code/autoresearch"><img src="https://agentmods.dev/badge/commands/drivelineresearch/autoresearch-claude-code/autoresearch/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for autoresearch

Your own site · 80×15
<a href="https://agentmods.dev/commands/drivelineresearch/autoresearch-claude-code/autoresearch"><img src="https://agentmods.dev/badge/commands/drivelineresearch/autoresearch-claude-code/autoresearch.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 14 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 544 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00014 $0.00544
Opus 5 $0.00007 $0.00272
Sonnet 5 $0.00003 $0.00109
Haiku 4.5 $0.00001 $0.00054

Measured yesterday against content hash c4f2fbbdbf67, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-09, from the pricing page.

Security

Grade A, and why

autoresearch scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

commands/autoresearch.md · 48 lines

What it actually says

Autoresearch command

Arguments: $ARGUMENTS

Read ${CLAUDE_PLUGIN_ROOT}/skills/autoresearch/SKILL.md and follow its protocol. Claude substitutes that plugin path in this command's text; it need not exist as an environment variable in Bash. For a manual install, where the placeholder is not expanded, read ~/.claude/skills/autoresearch/SKILL.md instead. Resolve AR_SCRIPTS to the scripts directory beside the file you read. Do not invoke autoresearch through the Skill tool again: this adapter shares that name and can shadow the shared skill. If neither known path exists, pause and report the missing installation rather than searching outside those locations.

Handle the requested mode before any resume or setup action:

  • off: create .autoresearch-off in the experiment workspace, preserve partial work, and stop. Do not launch another experiment.
  • status: read state/dashboard/recent worklog and show current segment budget, baseline/best, counts, and pause status. Do not resume or change experiment data. For a live Claude session, first create .autoresearch-inspect so the Stop hook permits this inspection turn to end. This small control write is the only write.
  • report: for a live Claude session create .autoresearch-inspect; write autoresearch-report.md with results, evidence limits, winning changes, failures, and remaining ideas. Do not run experiments or unpause. This mode writes a report.
  • resume (session exists): read session/state/worklog and git status/log. Resolve partial/unlogged work first and check the budget. Remove .autoresearch-off only when explicitly resuming within budget; do not reset the config to gain runs. An exhausted budget needs a user extension or deliberate new experiment contract.
  • fresh goal (no session): verify git and preserve existing work, then follow skill setup. Infer goal/metric/scope from supplied arguments where possible.

A user stop, cancellation, or new scope overrides continuation. Complete one experiment per autonomous turn; the Claude Stop hook handles the next turn when valid state and budget permit it. Never treat a plugin code review as a request to start an optimization session.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday Changed · -15 lines · +5 tokens per session c4f2fbbdbf67
  2. 10d ago First seen · 63 lines · 9 tokens per session scan A 7a7404d4ab9f

Subscribe to this mod's changes

autoresearch is a command published in the GitHub repository drivelineresearch/autoresearch-claude-code (342 stars, last pushed yesterday), licensed MIT. It adds 14 tokens to every session and 544 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.