DriveMind

English · 简体中文 · GitHub Repository · Issues

A calm reliability layer for AI agents.

Thoughtful execution, clear safety boundaries, and human collaboration for real work.

DriveMind is for people who want agents to work like careful collaborators, not reckless operators. It helps agents stay with meaningful work, keep humans informed, stop at unclear boundaries, and leave behind reusable lessons after the task is done.

Current release: v0.4.0

DriveMind v0.4.0 builds on the v0.3 decision layer and adds context intelligence: cross-session continuity, confidence signaling, and a zero-config installer across Claude Code, Codex, and OpenClaw.

This repository ships the first public DriveMind package: the skill, one-click installers, operating references, templates, examples, governance docs, and brand assets needed to use it like a real product.

What DriveMind stands for

DriveMind is a gentler and more rational form of reliability. It is built around three commitments:

1. continue with evidence, not stubbornness
2. escalate when the safety boundary becomes unclear
3. turn finished work into reusable memory, not disposable output.

Why DriveMind exists

Long-running agent work usually breaks in four ways:

1. context gets lost
2. work stops too early
3. the agent waits passively instead of helping
4. uncertainty turns into bad tradeoffs.

DriveMind exists to make that behavior calmer, safer, and easier to trust.

What ships in this repo

DriveMind v0.3.0 includes:

• a reusable DriveMind skill in skill/SKILL.md
• persistence, escalation, mode, task-typing, decision-gates, and stuck-resolution references in skill/references/
• diary, review, and distillation templates in skill/templates/
• local and remote install helpers in scripts/
• product, licensing, and operating docs in docs/
• contributor and governance docs at the repository root
• a minimal validation workflow in .github/workflows/
• first-party logo assets in assets/logo/
• practical walkthroughs and comparisons in examples/.

Host and validation status

DriveMind currently has three validated host/workflow surfaces:

• Claude Code
• Codex
• OpenClaw

Current install documentation in this repository is focused on Claude Code and Codex targets. OpenClaw is already a validated host environment for DriveMind, even though its usage model is closer to skill hosting and workflow execution than a one-click bootstrap target in this repo.

Install

One command — works automatically

The installer detects which AI tools you have installed (Claude Code, Codex) and installs to all of them. No flags needed.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.ps1 | iex

That's it. The installer finds your tools and installs DriveMind to each one. If nothing is detected, it defaults to ~/.claude/skills.

Install to a specific target

Advanced: choose a specific install target

macOS / Linux

# Claude Code personal
curl -fsSL https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.sh | bash -s -- claude-personal

# Codex personal
curl -fsSL https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.sh | bash -s -- codex-personal

# Claude project-local
curl -fsSL https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.sh | bash -s -- claude-project /path/to/your-project

Windows (PowerShell)

$tmp = New-TemporaryFile; Invoke-WebRequest https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.ps1 -OutFile $tmp; & $tmp claude-personal; Remove-Item $tmp
$tmp = New-TemporaryFile; Invoke-WebRequest https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.ps1 -OutFile $tmp; & $tmp codex-personal; Remove-Item $tmp
$tmp = New-TemporaryFile; Invoke-WebRequest https://raw.githubusercontent.com/Yuzc-001/DriveMind/main/scripts/bootstrap.ps1 -OutFile $tmp; & $tmp claude-project D:\path\to\your-project; Remove-Item $tmp

From a cloned repo

./scripts/install.ps1 codex-personal
./scripts/install.ps1 claude-personal
./scripts/install.ps1 claude-project D:\path\to\your-project

bash ./scripts/install.sh codex-personal
bash ./scripts/install.sh claude-personal
bash ./scripts/install.sh claude-project /path/to/your-project

See docs/installation.md for the full install matrix, aliases, update flow, and target paths.

Distribution model

DriveMind does not ship separate personal and project packages. It ships one portable drivemind skill package with different install targets.

That keeps the product coherent, easier to maintain, and easier to trust. See docs/distribution-strategy.md.

What DriveMind changes

DriveMind adds four operating behaviors to agent work:

• calm persistence: keep working on important tasks without turning every obstacle into drama
• clear safety boundaries: ask before crossing unclear or high-risk lines
• human collaboration: keep the human informed and return decisions at the right moment
• review and reusable memory: distill what happened so the next run starts smarter.

Who it is for

DriveMind is built for:

• developers shipping agent workflows
• advanced AI users running recurring tasks
• researchers exploring agent reliability
• teams that want calmer human-in-the-loop collaboration.

What it is not

DriveMind is not:

• a hype persona or roleplay wrapper
• a diary-only tool
• a memory plugin by itself
• a permission slip for reckless retries.

It is a reliability layer that makes agents steadier, more legible, and easier to supervise.

Basic activation

Examples:

• Enable DriveMind in execution mode.
• Use DriveMind for this task. Stay steady and keep me informed.
• Stay with this a little longer before concluding, but ask me before crossing an unclear boundary.
• Review this task after completion and preserve what should be remembered.

When DriveMind is active, the agent should:

1. gather evidence before concluding failure
2. try bounded alternatives before escalation
3. surface boundary questions clearly and early
4. leave behind reusable lessons after meaningful work.

Repository map

• README.md / README.zh-CN.md: public entry points in English and Chinese
• PROJECT.md: product definition and scope
• PROMO.md: launch and release copy
• CHANGELOG.md, CONTRIBUTING.md, TRADEMARKS.md, SECURITY.md, CODE_OF_CONDUCT.md: governance, release history, and project boundaries
• .github/workflows/: validation workflow for scripts and markdown links
• .github/ISSUE_TEMPLATE/: issue entry guidance and contact links
• �ssets/logo/: official logo system and source assets
• �ssets/social/: GitHub social preview asset and release-ready repository artwork
• docs/: brand kit, installation, licensing, principles, activation, interaction model, roadmap, and supporting product docs
• scripts/: local installers, remote bootstrap entry points, and reproducible logo export tooling
• skill/: the DriveMind skill, references, and templates
• examples/: walkthroughs that show DriveMind in practice.

Roadmap

v0.1 Foundation

• stable positioning and brand voice
• first reusable DriveMind skill
• one-click install for Codex and Claude Code
• portable personal and project install modes
• a standard source-available licensing model
• first logo and brand assets
• persistence, escalation, and collaboration references
• review, diary, and distillation templates
• examples, bilingual README, governance docs, and validation workflow.

v0.2 Validated

• stronger natural-language triggering for calm persistence and boundary-aware collaboration
• Intensive mode naming and clearer reporting expectations
• structured-but-adaptive review behavior
• review style guidance for more natural retrospectives
• validation across explicit triggers, implicit triggers, escalation, troubleshooting, judgment, and post-task review

v0.3 Formal Release

• lightweight task typing for judgment, boundary, progress, diagnosis, and distillation tasks
• explicit decision gates for high-impact actions, production/release decisions, external representation, and data/deletion work
• stuck-resolution guidance so persistence becomes blocker diagnosis and momentum recovery instead of brute repetition
• compression rule to keep short answers natural and avoid visible framework over-expansion

v0.4 Context Intelligence

• session handoff protocol for cross-session continuity: passive by default, triggered by user signals, invisible on short tasks
• confidence signaling reference: evidence-based / inference / hypothesis levels with verification paths
• language consistency: all skill references translated to English
• zero-config installer: auto-detects Claude Code and Codex, installs to all found targets with one command

Usage boundary

DriveMind is currently source-available under BUSL-1.1, not Open Source. Production use before March 11, 2030 requires separate permission from the licensor. The repository is scheduled to convert to GPL-2.0-or-later on March 11, 2030. See LICENSE.md and docs/licensing.md.

Collaboration & Contact

• Issues: https://github.com/Yuzc-001/DriveMind/issues
• Email: zxyu24@outlook.com

Use Issues for bugs, install problems, documentation gaps, and feature discussion. Use email for licensing, private collaboration, or questions that should not start in a public thread.

Star History

Read next

• README.zh-CN.md
• PROJECT.md
• CHANGELOG.md
• CONTRIBUTING.md
• TRADEMARKS.md
• PROMO.md
• docs/github-presence.md
• docs/brand-kit.md
• docs/licensing.md
• docs/installation.md
• docs/distribution-strategy.md
• docs/principles.md
• docs/interaction-model.md
• docs/activation.md
• docs/use-cases.md
• docs/with-vs-without.md
• docs/release-notes-v0.1.md
• skill/SKILL.md