oleg pustovit
isolated config layers for Claude Code, Codex, and OpenCode
Last verified:
What is oleg pustovit?
oleg pustovit is a declarative, safe control plane for deploying coding-agent assets across multiple AI coding tools. It manages and deploys AGENTS.md instructions, Agent Skills (SKILL.md), Claude Code slash commands (.claude/commands), and Codex custom prompts (~/.codex/prompts) to their respective tool-specific discovery locations. The tool creates an ephemeral configuration layer that allows teams to maintain reproducible AI coding environments without modifying their entire home directory.
Key features include cross-tool consistency (deploying the same assets to Codex, Claude Code, Cursor, VSCode via explicit target adapters), reusable and rollbackable deployments with preview/diff-first workflow, snapshots and rollback capabilities, and manifest-based safe deletes that protect user-owned files. It supports multi-machine sync through git with sync --rebase, treats the config repo as a single source of truth, and provides stable automation contracts through --json output and MCP integration.
oleg pustovit is designed for developers, engineering teams, and DevOps practitioners who use AI coding agents like Claude Code, OpenAI Codex, and OpenCode. It's particularly valuable for teams who need consistent agent configurations across multiple machines, want version-controlled agent skills, need to preview changes before applying them, and require rollback capabilities for safe configuration management. The tool follows a compiler model where knowledge docs are source files, skills are compiled artifacts, and packages are distributable output.
The workflow involves initializing a project with agentpack init, updating assets with agentpack update, previewing changes with agentpack preview --diff, deploying with agentpack deploy --apply, checking status with agentpack status, and rolling back with agentpack rollback --to <snapshot_id> when needed.
oleg pustovit pricing
Pricing model: Freemium
Agentpack is completely free and open source under the MIT license. It can be installed via Cargo with cargo install agentpack --locked or from source with cargo install --git https://github.com/liqiongyu/agentpack --tag v0.9.1 --locked. Prebuilt binaries are available on GitHub Releases at https://github.com/liqionnyu/agentpack/releases. There are no paid tiers, subscriptions, or premium features - all functionality including cross-tool deployment, rollback, git sync, JSON API, and MCP integration is available for free.
oleg pustovit pros
- Declarative configuration for predictable deployments
- Safe rollback with snapshot restoration
- Preview/diff before applying changes
- Cross-tool consistency across Codex, Claude Code, Cursor, VSCode
- Git-based sync with rebase support for multi-machine setups
- Single source of truth via config repository
- Manifest-based safe deletes protect user files
- JSON API for automation and scripting
- MCP integration for orchestration
- Open source under MIT license
- Small footprint at 1MB with 19K SLoC
- Explicit target adapters and mappings
- Event logging for auditability
- Layered overlays (global/machine/project)
- Stable automation contracts for CI/CD integration
oleg pustovit cons
- Rust-based requiring Cargo for installation
- Focused only on agent assets not entire home directory
- Learning curve for manifest syntax
- Requires git repository for multi-machine sync
- Prebuilt binaries may not be available on crates.io
- Unstable releases (2 unstable versions as of Jan 2026)
- Target-specific discovery locations can be complex
- No built-in GUI for configuration
- Automation requires --json --yes flags
- Documentation split across English and Chinese
- Only supports specific tools (Codex, Claude Code, OpenCode)
- Lockfile management adds complexity
- No cloud hosting - purely local solution
- Nested dependency tracking for skills
- CI integration requires manual wiring
Frequently asked questions about oleg pustovit
What is Agentpack and what problem does it solve?
Agentpack is a declarative, safe control plane for deploying coding-agent assets across tools. It solves the problem of maintaining reproducible AI coding environments by managing AGENTS.md instructions, Agent Skills (SKILL.md), Claude Code slash commands, and Codex custom prompts. Instead of manually configuring each tool, Agentpack provides an ephemeral configuration layer that ensures consistent agent behavior across multiple machines and tools.
Which AI coding tools does Agentpack support?
Agentpack explicitly supports Codex, Claude Code, Cursor, VSCode, and OpenCode through target adapters and mappings. It deploys assets to tool-specific discovery locations like ~/.codex/prompts for Codex, .claude/commands for Claude Code, and .opencode/skills for OpenCode.
How do I install Agentpack?
Install via Cargo with cargo install agentpack --locked. If crates.io install is not available, install from source using cargo install --git https://github.com/liqiongyu/agentpack --tag v0.9.1 --locked. Prebuilt binaries are also available on GitHub Releases.
What is the typical workflow for using Agentpack?
The end-to-end workflow is: agentpack init to initialize, agentpack update to fetch assets, agentpack preview --diff to see changes, agentpack deploy --apply to apply changes, agentpack status to check current state, and agentpack rollback --to <snapshot_id> to revert if needed. Always run preview before deploy in automation.
How does rollback work in Agentpack?
Agentpack creates snapshots every time you deploy with deploy --apply. The snapshot_id is returned in the JSON output (data.snapshot_id). You can restore any previous state using agentpack rollback --to <snapshot_id>. Snapshots are stored in state/snapshots/ directory.
Why not use a dotfiles manager like Stow, chezmoi, or yadm?
Agentpack is focused specifically on deploying agent assets into tool-specific discovery locations, not managing your entire $HOME directory. Dotfiles managers handle general configuration files, while Agentpack provides specialized support for AI coding agent assets with features like preview/diff, rollback, and cross-tool consistency that dotfiles managers don't offer.
How do I sync Agentpack configurations across multiple machines?
Treat the config repository as a single source of truth and use agentpack sync --rebase to sync configurations across machines. This leverages git for version control and conflict resolution, ensuring all machines have consistent agent configurations.
What is the JSON API and when should I use it?
The JSON API provides machine-readable output for automation. Use --json flag with commands like deploy --apply and preview. In automation, prefer --json --yes and always run preview first. The deploy command returns data.snapshot_id in JSON which you can pass to rollback commands.
What are Agent Skills and how do they work?
Agent Skills are SKILL.md files that define agent behavior and knowledge. Agentpack manages these skills, tracking which source docs each skill depends on, warning when sources change, and validating everything before publishing. Skills are materialized into .claude/skills/ and .agents/skills/ where agent runtimes discover them at runtime.
How does Agentpack handle multi-machine configuration consistency?
Agentpack uses a config repository as a single source of truth, synchronized via git with sync --rebase. It supports layered overlays (global, machine-specific, and project-level) that compose together. The lockfile (agentpack.lock.json) ensures deterministic builds, and the manifest system tracks exactly what was deployed to each target.