Home

Home

dotclaude is an opinionated Claude Code plugin marketplace for software engineering. It keeps only the parts that Claude Code does not cover, and it optimizes for the most quality per unit of usage quota, not for speed. Since 0.27 the core plugin uses only official extension points: settings, permission rules, the sandbox, one output style, agents, skills, and one hooks module. If Claude does something because of dotclaude and you ask "why did you do that?", the answer is in these pages.

What it does

• Gives permission rules and a sandbox in the settings snippet, so that Claude Code asks before destructive, public, and secret-reading actions.
• Gives an ask rule for each recursive rm in the settings snippet (the plugin alone gives none), and asks before the first contact with a repository of another owner that has an AI policy file.
• Gives settings in the snippet that turn off compaction and set a 300K context window, so that a session stops at the limit and you start fresh with /clear. The plugin itself ships an empty settings.json.
• Forces one output style (role, tone, and format) and puts the working rules in the system prompt.
• Gives Claude five agents with a model, an effort, and a turn limit each (Agents).
• Works with OpenSpec for the task list, so work continues after /clear with the openspec-apply-change skill.
• Shows model, effort, context, the OpenSpec change, and the 5-hour limit in a status line.

The plugins

Each plugin has one reference page. The page says what the plugin does, how to install it, and what it holds: skills, bins, hooks, options, and limits. Each add-on works without the core plugin.

Plugin	What it does	Reference
dotclaude	The core: output style, system prompt rules, agents, guards, status line, and drift check.	Core plugin
dotclaude-codegraph	Gives Claude the code graph of a project from the codegraph CLI.	Code graph
dotclaude-browser	Drives a browser and answers WebSearch in the headless browser.	Browser
dotclaude-jev	A second opinion on a decision from TypeSafe Jev.	Second opinion
dotclaude-docs	Current library docs from Context7.	Library docs
dotclaude-sessions	Measures the start of a session and the usage of each run.	Sessions
dotclaude-lab	Tools for plugin authors: sandbox, capture, schema, and eval compare.	Lab

New in 0.28.0

• A system prompt variant for each model.
• A context bound for subagents.
• Semantic line breaks in written text, with no model token.
• A new plugin layout, and the plugins dotclaude-sessions and dotclaude-lab.
• Evals through claude plugin eval (Evals), with a suite of 16 cases in plugins/dotclaude/evals/ and just eval to run it.
• The decisions D1 to D35 with their evidence.
• Removed: the setup profile, the router, the handoff skill, usage state, the Bun eval harness and the root evals/ folder (design D35), and dotclaude-modder. The settings snippet replaces the profile.

The position

Position	What it means
Quality per quota over speed	Usage limits stop work. dotclaude spends quota where it buys quality, and it locks fast mode off.
Quality over quantity	One checked change is worth more than several unchecked changes. Claude reproduces a bug before it fixes it, and it reports what it did not check.
Official points over runtime code	When Claude Code has a setting, a rule, or a hook for a need, dotclaude uses it and writes no runtime code.
Evidence over habit	Each bound comes from a measurement, an official source, or a reverse-read of the Claude Code binary.
The user decides	Rules ask before hard-to-reverse or public actions, and they never approve anything. The setup skill shows every step before it runs.

Where to read

Page	Read it to
Quickstart	See what happens in a first session.
Install	Install the marketplace and the plugins.
Settings snippet	Copy the keys that Claude Code needs. See also snippet keys and schema.
Hooks	See each hook of each plugin, its event, and what it does.
Troubleshooting	Fix a symptom.
Undocumented reads	See what dotclaude reads that Claude Code does not document, and how it breaks.
Guards	Learn what asks, what denies, and what the sandbox does.
Development and Sandbox	Change dotclaude and test a checkout.
Attributions	See the projects whose ideas dotclaude reimplements.

Concepts and evidence

Page	What it holds
Design	The 0.27 rebuild, design principles, and rejected alternatives.
Decisions	The design decisions D1 to D35.
Usage evidence	Where one Max 20x week of usage went, and what that means on Pro.
Plans and models	Prices, model fit, and effort.
Prompt surface	What a request contains, the simple prompt, and the forced style.
Eval history	The eval runs of the 0.4 to 0.28 releases.
Claude mods	The plugin hooks modules of Claude Code, and the gaps.
Open items	Work still to measure or decide.
Release history	What changed in each older release, and why.

The CHANGELOG lists the current release.

Source labels

The evidence pages give the proof for each choice, and each fact there has one of these labels. A choice with no label is the maintainer's decision, and the page says so.

Label	Source
official	Anthropic documentation or support pages
binary	read from the installed Claude Code bundle, never patched
capture	a request body that Claude Code sent to a local listener
measured	a run or a transcript scan on the maintainer's machine
reported	user reports, weighed by score and agreement
inference	a conclusion from the facts above

The 0.27 facts date from Claude Code 2.1.292. The older evidence dates from 2026-09-26 to 2026-10-04 and Claude Code 2.1.283 to 2.1.289. Dollar figures are API-equivalent list prices. How a subscription weights cache reads against output is not published.