Search everything...

Stats

Actions

Available In

autoresearch

Name: autoresearch
Author: hironow

By hironow

Autonomous experiment loop plugin inspired by karpathy/autoresearch. Iteratively modifies target code, evaluates against fixed metrics, and keeps only improvements via git-based keep/revert cycles.

npx claudepluginhub hironow/dotfiles --plugin autoresearch

Popularity

Stars

Med: 0·Avg: 285

Installs

Med: 0·Avg: 1

What's Inside

Agents1

researcher

/researcher

Use this agent when running autonomous experiment iterations that modify target code, evaluate against fixed metrics, and keep or revert based on results. This agent handles a single experiment cycle independently. <example> Context: The user has set up an experiment config and wants to start optimizing. user: "Start running experiments to optimize the algorithm performance" assistant: "I'll launch the researcher agent to begin the autonomous experiment loop." <commentary> The user wants autonomous experimentation. The researcher agent handles the modify-evaluate-decide cycle independently. </commentary> </example> <example> Context: An experiment loop is in progress and the user wants to continue. user: "Keep running more experiments on the current branch" assistant: "I'll spawn the researcher agent to run the next experiment iteration." <commentary> Continuing an existing experiment loop. The researcher reads current state from git and results.tsv to pick up where it left off. </commentary> </example> <example> Context: The user wants to try a specific optimization idea within the loop. user: "Try increasing the batch size and see if it helps" assistant: "I'll launch the researcher agent with that specific hypothesis." <commentary> User provides a specific hypothesis. The researcher implements, evaluates, and decides keep/revert based on the result. </commentary> </example>

Skills4

analyze-results

/analyze-results

Analyze results from an autoresearch experiment loop. This skill should be used when the user asks to "analyze experiment results", "show research progress", "summarize experiments", "what improved", "experiment statistics", or needs to review and visualize the results from results.tsv.

research-loop

/research-loop

Core methodology reference for the autoresearch keep/revert cycle. This skill provides the rules, decision logic, and error recovery patterns for the autonomous loop — not the top-level entry point (use /research for that). Applicable when the user asks about "autoresearch pattern", "keep/revert logic", "experiment decision criteria", "auto-tune hyperparameters", "run keep/revert optimization", or needs guidance on how the autonomous modify-evaluate-decide cycle works internally.

research

/research

This skill should be used when the user asks to "start research", "run autonomous experiments", "optimize my code automatically", "begin experiment loop", "run autoresearch", or needs to start or resume an autonomous research experiment loop. This is the top-level entry point that orchestrates setup and loop execution. For internal keep/revert methodology, see the research-loop skill.

setup-experiment

/setup-experiment

This skill should be used when the user asks to "set up an experiment", "initialize a research run", "create experiment config", "prepare experiment branch", or needs to configure the autoresearch experiment environment before starting an autonomous optimization loop.

Hooks1

Event Hooks

File writes

1 hook across 1 event

Stats

Version0.1.0

LanguagePython

Stars0

MaintenanceGood

LicenseMIT

Last CommitMar 21, 2026

AddedMar 25, 2026

Actions

View on GitHub View README Plugin Marketplace JSON

Own this plugin?

Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).

Available In

hironow-plugins

Safety Signals

Caution

Modifies files

Hook triggers on file write and edit operations

Uses power tools

Uses Bash, Write, or Edit tools

README

dotfiles

My configuration for Zsh, Mise, Just, and more.

Installation

Requires curl and git.

bash -c "$(curl -fsSL https://raw.githubusercontent.com/hironow/dotfiles/main/install.sh)"

[!NOTE]
Mac, Linux, Windows(WSL内Linux)へ対応

usage

# just (task runner)
just help

just sync-agents-preview
just sync-agents

just update-all
just dump

# diagnostics
just self-check
# run with quick validate tests (needs Docker)
just self-check with_tests=1
just doctor
just validate-path-duplicates

# uv on mise
mx uv sync

# just on mise
mx just --list

# mise env
mx dotenvx run -- mise set
# mise env with github credentials (use gh extension)
gh do -- mise set

# set env by dotenvx (encrypted)
mx dotenvx set HELLO World
# set env by mise (plain, unencrypted)
mx mise set WORLD=hello

tests (docker required)

# run all sandbox tests
just test

# run by pytest marker (install/validate/versions/deploy/check)
just test-mark marker=validate

# verify install.sh (docker required)
just test-install

install options

# full install
bash ./install.sh

# lightweight (skip heavy tools)
INSTALL_SKIP_HOMEBREW=1 INSTALL_SKIP_GCLOUD=1 INSTALL_SKIP_ADD_UPDATE=1 bash ./install.sh

setup for https localhost

# check A record for localhost -> 127.0.0.1
dig localhost.hironow.dev

# create/update cert for https
sudo certbot certonly --manual --preferred-challenges dns -d localhost.hironow.dev --config-dir ${config_root}/private/certificates

# check simple-server for https localhost
cd tools/simple-server
sudo mise x -- go run main.go

Tools

Tools: Collection of utility scripts and tools (e.g., RTTM converter, simple server).

references

mcp setup

claude mcp add -s user chrome-devtools bunx chrome-devtools-mcp@latest
claude mcp add -s user -t http deepwiki https://mcp.deepwiki.com/mcp

# latest documentation MCP per user
claude mcp add -s user -t http bun https://bun.com/docs/mcp
claude mcp add -s user -t http cloudflare https://docs.mcp.cloudflare.com/mcp
claude mcp add -s user -t http vercel https://mcp.vercel.com
claude mcp add -s user -t http livekit-docs https://docs.livekit.io/mcp
claude mcp add -s user -t http openai https://developers.openai.com/mcp
# -- Google Cloud: https://developers.google.com/knowledge/mcp#gcloud-cli
YOUR_PROJECT_ID=<your-project-id>
gcloud beta services mcp enable developerknowledge.googleapis.com --project=$YOUR_PROJECT_ID
# -- needs credentials setup
# gcloud auth login
# gcloud auth application-default login
gcloud services api-keys create --project=$YOUR_PROJECT_ID --display-name="DK API Key"
YOUR_API_KEY=<your-keyString>
claude mcp add google-dev-knowledge -s user -t http https://developerknowledge.googleapis.com/mcp --header "X-Goog-Api-Key: $YOUR_API_KEY"
# -- AWS: https://awslabs.github.io/mcp/servers/aws-knowledge-mcp-server/
claude mcp add -s user -t http aws-knowledge-mcp-server https://knowledge-mcp.global.api.aws
# -- k6: https://grafana.com/docs/k6/latest/release-notes/v1.6.0/#introducing-mcp-k6-ai-assisted-k6-script-writing-mcp-k6
claude mcp add --scope=user --transport=stdio k6 -- docker run --rm -i grafana/mcp-k6

# specific (needs copy for other agents' directory) per project
claude mcp add -s project -t http jaeger http://localhost:16687/mcp

MCP catalog refs.

skill setup

justfile経由で CLAUDE_CONFIG_DIR を切り替えつつ操作できる:

just skills ls                # デフォルトconfigでスキル一覧
just skills add <repo> --all  # スキル追加
just env=a skills ls -g       # ~/.claude-work-a 向け
just env=b skills ls -g       # ~/.claude-work-b 向け
just env=c skills ls -g       # ~/.claude-work-c 向け
just env=d skills ls -g       # ~/.claude-work-d 向け
just env=p skills ls -g       # ~/.claude (personal) 向け

just skills add vercel-labs/agent-skills
just skills add modelcontextprotocol/ext-apps
just skills add wandb/skills
just skills add https://github.com/googleworkspace/cli
# browser: https://github.com/vercel-labs/agent-browser?tab=readme-ov-file#agentsmd--claudemd
just skills add vercel-labs/agent-browser

Skill catalog refs.

https://skills.sh/

git setup

# avoid merge commits when pulling
git config --global pull.rebase true

win setup (WSL)

View full README on GitHub

autoresearch

Popularity

What's Inside

Confidence

README

dotfiles

Installation

usage

tests (docker required)

install options

setup for https localhost

Tools

references

mcp setup

skill setup

git setup

win setup (WSL)

Similar Plugins

dotfiles-commands

mcp-tools

b00t

dev-environment

ai-driven-development

geepers

More by hironow

autodesign

dotfiles

Installation

usage

tests (docker required)

install options

setup for https localhost

Tools

references

mcp setup

skill setup

git setup

win setup (WSL)

More by hironow

autodesign

Popularity

Health & Quality

Similar Plugins

dotfiles-commands

mcp-tools

b00t

dev-environment

ai-driven-development

geepers