[PR #268] Add Dao Code

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/awesome-deepseek-agent/pull/268 **Author:** [@tigicion](https://github.com/tigicion) **Created:** 6/29/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `add-dao-code` --- ### 📝 Commits (1) - [`22270ea`](https://github.com/deepseek-ai/awesome-deepseek-agent/commit/22270ea5d3e3b579165b2ff390b87554a4df2034) Add Dao Code integration guide ### 📊 Changes **4 files changed** (+226 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+1 -0) 📝 `README.zh-CN.md` (+1 -0) ➕ `docs/dao_code.md` (+112 -0) ➕ `docs/dao_code.zh-CN.md` (+112 -0) </details> ### 📄 Description ## Add Dao Code Adds a bilingual integration guide for **Dao Code** — an open-source (MIT) TypeScript terminal AI coding agent built for DeepSeek-V4. - Repo: https://github.com/tigicion/dao-code - npm: `dao-code` (command `dao`) ### What it is A terminal-native coding agent engineered around DeepSeek-V4's cache economics: a prefix-cache hit costs ~1/120 of a miss, so Dao Code keeps the system prefix, tool table and memory byte-stable to maximize the hit rate and runs reflection/memory on forks that reuse the main prefix cache — letting cross-session memory and a challenger/refocuser self-correction layer add almost no token cost (95.8% measured cache hit rate over 3.89M input tokens on 7 real OSS bug-fix tasks). Its distinctive feature is cross-session memory that is deterministically re-verified against the live codebase on startup, so stale facts are pruned rather than accumulated. Full 1M context, a security sandbox with layered `allow`/`ask`/`deny` permissions, Agent Skills, MCP (stdio + HTTP/SSE), hooks, sub-agents, and an autonomous long-task mode (`/goal`). Config-compatible with Claude Code (`settings.json` / `SKILL.md` / `hooks.json` / `mcp.json`). ### Files changed - `docs/dao_code.md` (English guide) - `docs/dao_code.zh-CN.md` (Simplified Chinese guide) - `README.md` — added table row (alphabetical order, after Crush / before Deep Code) - `README.zh-CN.md` — added table row (same position) ### Agent Configuration - [x] Model names are current (v4-pro / v4-flash, not v3 names) - [x] 1M context is configured or mentioned - [x] Max thinking effort level is supported - [x] Pricing is current (input, output, cache hit), verified against DeepSeek API Docs - [x] Config fields are verified to work in the agent - [x] No workarounds for already-fixed upstream bugs ### Documentation - [x] Both English and Chinese versions included in a single PR - [x] README table entry inserted in alphabetical order - [x] One tool per PR; unrelated tools not combined - [x] Images placed in `docs/assets/` with descriptive filenames and reasonable sizes (n/a — no images) --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi commented

2026-10-26 09:47:45 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/awesome-deepseek-agent/pull/268
Author: @tigicion
Created: 6/29/2026
Status: 🔄 Open

Base: main ← Head: add-dao-code

📝 Commits (1)

22270ea Add Dao Code integration guide

📊 Changes

4 files changed (+226 additions, -0 deletions)

View changed files

📝 README.md (+1 -0)
📝 README.zh-CN.md (+1 -0)
➕ docs/dao_code.md (+112 -0)
➕ docs/dao_code.zh-CN.md (+112 -0)

📄 Description

Add Dao Code

Adds a bilingual integration guide for Dao Code — an open-source (MIT) TypeScript terminal AI coding agent built for DeepSeek-V4.

Repo: https://github.com/tigicion/dao-code
npm: dao-code (command dao)

What it is

A terminal-native coding agent engineered around DeepSeek-V4's cache economics: a prefix-cache hit costs ~1/120 of a miss, so Dao Code keeps the system prefix, tool table and memory byte-stable to maximize the hit rate and runs reflection/memory on forks that reuse the main prefix cache — letting cross-session memory and a challenger/refocuser self-correction layer add almost no token cost (95.8% measured cache hit rate over 3.89M input tokens on 7 real OSS bug-fix tasks). Its distinctive feature is cross-session memory that is deterministically re-verified against the live codebase on startup, so stale facts are pruned rather than accumulated. Full 1M context, a security sandbox with layered allow/ask/deny permissions, Agent Skills, MCP (stdio + HTTP/SSE), hooks, sub-agents, and an autonomous long-task mode (/goal). Config-compatible with Claude Code (settings.json / SKILL.md / hooks.json / mcp.json).

Files changed

docs/dao_code.md (English guide)
docs/dao_code.zh-CN.md (Simplified Chinese guide)
README.md — added table row (alphabetical order, after Crush / before Deep Code)
README.zh-CN.md — added table row (same position)

Agent Configuration

Model names are current (v4-pro / v4-flash, not v3 names)
1M context is configured or mentioned
Max thinking effort level is supported
Pricing is current (input, output, cache hit), verified against DeepSeek API Docs
Config fields are verified to work in the agent
No workarounds for already-fixed upstream bugs

Documentation

Both English and Chinese versions included in a single PR
README table entry inserted in alphabetical order
One tool per PR; unrelated tools not combined
Images placed in docs/assets/ with descriptive filenames and reasonable sizes (n/a — no images)

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

bsakaguchi added the

pull-request

label

bsakaguchi referenced this issue

2026-10-26 09:48:01 +00:00

[PR #300] fix: update Cherry Studio Chinese website URL to cherryai.com.cn #324