Shadowclone

Shadowclone maintains the skills, instructions, and checks that help Claude Code, Codex, Pi, Cursor, and Antigravity follow your engineering preferences. It works with the coding agents you already use and keeps you in control of what it reads and changes.

Get started

Let your agent set it up

In Claude Code, add and install the plugin:

/plugin marketplace add theonly1me/shadowclone
/plugin install shadowclone@shadowclone

In Codex, add the marketplace and install Shadowclone from the Plugins Directory:

codex plugin marketplace add theonly1me/shadowclone
codex plugin add shadowclone@shadowclone

Then tell the agent: Set up Shadowclone. The packaged setup skill checks the CLI version, offers installation or an upgrade when needed, asks you for three separate consent choices, installs native guidance for detected agents, and prints the local build URL. This setup flow requires CLI 0.0.13 or newer. It never uses elevated privileges or chooses data access for you.

Cursor, Antigravity, and other skill-compatible agents can use the portable setup-shadowclone skill. Once any supported agent completes setup, Shadowclone synchronizes native guidance across the detected agents on your machine.

Set it up manually

Install the CLI:

npm install -g @shadowclone/cli

Step	Command	What happens	Time
1	`shadowclone init`	Answer three questions about learning and skill maintenance	About 1 minute
2	`shadowclone learn --deep`	Learn from enabled past sessions; repeat if more history remains	Up to 5 minutes per run
3	`shadowclone learning pending` and `shadowclone skills pending`	Review learned rules, scope, and publication before relying on them	About 1 minute

Setup takes about two to three minutes, excluding learning. Deep learning needs an installed and configured claude, codex, cursor-agent, or pi CLI. Pi setup uses any suitable model configured in Pi, including local providers. Claude Code learning on Linux needs bwrap and socat for its sandbox.

Skip learning. Run shadowclone wizard by itself if you do not want Shadowclone to read sessions or call a model. It starts from an empty active environment and lets you choose a build locally.

Run shadowclone to see the current result and next action. Use shadowclone wizard to browse optional workflow skills in a browser.

Delegate a pull request. The optional shadowclone-work skill takes a request, an issue, or an existing PR to ready for review: green checks, conflicts resolved, and review comments fixed or left for you. It replies to fixed comments with commit SHAs only and never merges. Use delegated work.

Run a cloud bot. Ask your agent to set up Shadowclone cloud, or run shadowclone bot setup --bot <login> in a repository checkout. The bot is a GitHub account that you name. You add its token and your Claude token on GitHub, and it reads your skills from a private repository. It works on issues, answers mentions, and reviews pull requests. Live installation qualification is pending. Set up a cloud bot.

Evaluations

We tested 24 fixed engineering tasks across eight preference families, with three repetitions per setup and model. Scores measure preference adherence, with each family weighted equally.

Setup	GPT 6.1 Sol	GPT 6 Luna	Claude Sonnet 5.5	Claude Opus 5.5
Agent alone	77.3%	59.0%	56.5%	53.2%
Existing user skills	85.2%	80.6%	76.4%	73.6%
Existing skills + Shadowclone routing	86.1%	72.7%	76.9%	75.9%
Existing skills + handwritten preferences	92.4%	81.2%	91.7%	86.1%
Existing skills + Shadowclone learning	89.4%	83.3%	90.0%	82.9%

Shadowclone learning improved recorded adherence over existing skills on all four models. The gain ranges from 2.8 to 13.7 percentage points. Handwritten preferences provide a reference for what agents can follow when the intended rules are supplied directly.

The learning setup includes routing and actual learned preferences from synthetic correction sessions. These results measure preference delivery on this task set. Read the setup definitions, family scores, methods, and measurement limits.

Privacy comes first

There is no Shadowclone collection service or telemetry. Learning records, revisions, and original-library snapshots stay on your machine. Model work goes through the authenticated agent CLI you select and is subject to that provider's terms.

You choose what it reads. Every capture source is off by default and has its own consent setting. Reading sessions does not silently enable memory, repository metadata, or skill editing.

Operation	What the selected provider can receive
Learning and skill maintenance	Selected redacted instructions, steering, and supporting context
Personal GitHub clone	The reviewed skills, native rules, selected repository, and issue or PR task
Browser editor	Nothing merely from opening it; optional AI drafting sends reviewed form fields
Evaluation	Synthetic prompts, the tested setup's guidance, and generated code

Redaction is a boundary, not an anonymity guarantee. Tool results, tool-returned file contents, thinking blocks, and data-access results are excluded from learning. Use only sources and repositories you are authorized to send to the selected provider.

Removal stays in your control. Disable background learning with shadowclone learning disable, stop automatic skill edits with shadowclone skills automatic off, or remove recorded state with shadowclone forget --all. See the privacy policy, data handling, and enterprise controls for sources, storage, managed policy, and removal details.

Guides

Guide	Use it to
Agent builds	Choose, create, and equip skills
Learning	Run deep learning, background learning, or record a preference
Skill maintenance	Review updates, conflicts, history, and undo
Repository setup	Share checks and skills with a repository
How it works	Follow guidance from consented evidence to coding agents
Command reference	Find everyday and advanced commands

Further reading: Documentation index · Architecture · Design history · Migration · Contributing · Security

Shadowclone is MIT licensed.