Maintainer of this skill?
Automated security scan — not a certification.
DeepAgent Code: AI coding agent with persistent memory and control plane
# Add to your Claude Code skills
git clone https://github.com/deepagent-ltd/deepagent-codeGuides for using ai agents skills like deepagent-code.
Last scanned: 7/18/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-07-18T06:04:22.098Z",
"npmAuditRan": false,
"pipAuditRan": true,
"promptInjectionRan": true
}See how deepagent-code compares with popular alternatives.
deepagent-code is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by deepagent-ltd. DeepAgent Code: AI coding agent with persistent memory and control plane. It has 436 GitHub stars.
Yes. deepagent-code passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/deepagent-ltd/deepagent-code" and add it to your Claude Code skills directory (see the Installation section above).
deepagent-code is primarily written in TypeScript. It is open-source under deepagent-ltd on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh deepagent-code against similar tools.
No comments yet. Be the first to share your thoughts!
30 days in the Featured rail · terms & refunds
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
Automated security scan — not a certification.
DeepAgent Code (DAC) combines a desktop coding workspace and a terminal agent with a shared execution core. It helps you work through changes that span files, model turns, tools, and people: keep project context, maintain an editable plan, delegate independent tasks, inspect execution, and bring the final answer back to the conversation that started the work.
This source line targets 2.0.4. See GitHub Releases for available installation packages.
| What you need | How DAC helps |
|---|---|
| Keep a complex task moving | Persistent sessions, editable goals and plans, live steering, and queued instructions keep work organized across many turns. |
| Stop repeating project context | Inspectable project memory connects facts, decisions, skills, and execution evidence, with provenance and controlled admission into model context. |
| Split work without losing control | Subagents handle independent tasks with durable run records and isolated worktrees for write-capable collaboration. |
| Move between code and the web | A Desktop-managed browser brings page inspection and authorized interaction into the same task, including controlled file exchange. |
| Turn a discussion into real work | Public IM supports /dac commands to your own local DAC, continuous follow-up in the same Session, and final answers returned to the original conversation. |
| Choose your tools and models | GUI and TUI share the core; configurable providers, Skills, domain packs, and MCP extend it for your project. |
Personal agent in IM. Use /dac to ask your own DAC to work on a discussion. It uses the normal local Session, model, tools, and permissions. Commands and execution are private to you; the final answer returns to IM. Follow-up requests in the same conversation reuse the Session. IM files and images have a 10 MiB per-attachment limit.
Desktop-managed browser. Inspect pages, navigate tabs, and perform supported page interactions from the coding workspace. Uploads, downloads, clipboard access, and other sensitive operations have explicit authorization and target checks. Available actions depend on the host platform and its browser capabilities.
Traceable event workflows. Configured events can drive task graphs with dependencies, execution records, and results. The value is practical: track what triggered the work, which step owns it, what completed, and which outcomes need attention.
A more consistent workspace. Model and reasoning controls sit near the composer actions, collaboration settings live in Settings, and IM has its own model selection. Desktop builds share the Native asset preparation and package-validation path across build workflows.
See the changelog for the user-facing release details.
Session inputs are persisted before execution is scheduled. While DAC works, send guidance to steer it at the next safe provider-turn boundary, or queue a separate task to begin after the active task settles. Goals add an editable plan and recorded progress to longer jobs.
Persistence also makes uncertainty visible. After an interruption, DAC distinguishes recorded work from external operations whose outcome needs recovery; it does not assume an uncertain tool or provider call is safe to repeat.
Project knowledge carries types, scope, status, versions, and source references. Saved decisions and approved experience can be reused in later sessions; you can inspect and revise them rather than relying on hidden prompt history.
Context is assembled within a budget. Code references, project memory, execution documents, Skills, and domain knowledge are admitted as needed. Stable instructions and changing task state are kept separate to support provider caching, without making a fixed token-saving promise.
Delegate independent research, implementation, and review tasks to subagents. Durable task records preserve ownership, status, results, and parent notifications; write-capable collaboration uses isolated worktrees and review boundaries.
The event workflow layer adds dependency-aware coordination for configured automation. It complements interactive model execution with durable delivery and observable work state.
DAC combines file editing, terminal commands, LSP-backed code navigation, a managed browser, and project knowledge. Skills express repeatable procedures; domain packs add specialist knowledge; MCP connects external tools and resources through the same permission-aware workflow.
Choose an official provider or configure a compatible endpoint. Model and tool support depends on the selected provider's capabilities, and IM can use a separately selected model from your local configuration.
For example, while discussing a bug with a teammate:
/dac Inspect the login handler in this project, fix the failing case,
and run the relevant tests.
DAC opens or reuses the local Session associated with that conversation. View the streaming execution in the normal DAC workspace, then receive the final answer in the IM thread. Continue with another /dac request to build on the same Session; you can explicitly start a new Session when you want separate context.
IM activation is a separate, explicit purchase: 1 credit, valid indefinitely, with no automatic renewal. Connecting a model provider does not automatically activate or charge for IM. Obtain credentials through your administrator or an authorized distribution channel, activate IM in Settings, and choose the model for your personal agent.
Public IM transports messages; your agent executes on your own machine. DAC 2.0.4 does not provide a public remote-execution relay, shared agents callable by other users, or whole-desktop Computer Use.
Download the package for your platform from GitHub Releases. Open DAC, connect a model provider in Settings, choose a project, and start a conversation.
Use the Bun version pinned in the root package.json: Bun 1.4.2 for this source line.
git clone --branch 2.0.4 https://github.com/deepagent-ltd/deepagent-code.git
cd deepagent-code
bun install
From the repository root:
# Desktop workspace
bun run dev:desktop
# Terminal interface
bun run dev
# Web UI development server
bun run dev:web
Run a one-shot task or inspect an import before writing history:
bun run --cwd packages/deepagent-code dev run "Find and fix the failing login test"
bun run --cwd packages/deepagent-code dev import-history --from codex --dry-run
The standalone web UI needs a running DAC backend; the web UI guide shows the two-server setup.
See the Desktop package guide, web UI guide, and CLI/server guide for package-specific development and build commands.
In Settings → Providers, connect an available provider or add a compatible endpoint with its base URL and API key. Select the model and reasoning level for the task. The DeepAgent API Platform is one supported option; you can also bring your own provider credentials.
For the installed terminal application:
deepagent auth login
deepagent auth list
deepagent-code
On macOS and Linux, configuration and local state live under ~/.deepagent/code/. On Windows, configuration lives under %APPDATA%\deepagent-code and data under %LOCALAPPDATA%\deepagent-code. Provider credentials follow their configured authentication path. MCP secret handles can use native credential storage where available; credentials, tool permissions, and outbound actions retain their own execution boundaries.
DeepAgent Code is licensed under AGPL-3.0-or-later. See LICENSE for the full terms.
DAC builds on opencode and other open-source components. Original license notices and attribution are retain