v1.2.32 — Zed ACP Agent

Zed Editor Integration (ACP)

Codeep is now available as a native AI agent inside Zed via the Agent Client Protocol (ACP).

Setup

1. Install Codeep
```bash
npm install -g codeep
```

2. Add to Zed settings (`Cmd+,` → open `settings.json`)
```json
"agent_servers": {
"Codeep": {
"type": "custom",
"command": "codeep",
"args": ["acp"]
}
}
```

3. Open the Zed Agent panel and start chatting. Use `/help` to see all available slash commands.

---

What's new in 1.2.32

• `agent_thought_chunk` — thinking text is now sent as a separate thought update (not chat text)
• `tool_call` session/update notifications — Zed can display live tool activity (read/edit/delete/search/execute/fetch) with file locations
• Removed `Iteration X/100` from chat output (internal detail)

ACP features added since 1.2.26

• 30+ slash commands available in Zed autocomplete (`/commit`, `/diff`, `/scan`, `/review`, `/skills`, …)
• Workspace init on session start — creates `.codeep` folder, loads or creates session
• Session preview on load — shows last 6 messages
• `file/edit` notifications for structured file change tracking in Zed
• Skills execution (headless, streaming output to Zed chat)
• `/add` and `/drop` file context management

Codeep v1.2.18 — npm package fix

Fixes

• npm package cleanup — Compiled test files (*.test.js, *.test.d.ts) are no longer included in the published npm package. Package size reduced from 1.1MB (176 files) to 843KB (130 files).

Installation

npm install -g codeep

See v1.2.17 for all new features in this release cycle.

Codeep v1.2.17 — Multi-provider expansion, MCP tools & security hardening

What's New in v1.2.17

This release brings major provider expansion, MCP-powered agent tools, performance improvements, security hardening, and comprehensive test coverage.

---

New Providers

Provider	Models
DeepSeek	DeepSeek V3 (general), DeepSeek R1 (reasoning with chain-of-thought)
Anthropic	Claude Sonnet 4.6, Claude Opus 4.6, Claude Haiku 4.5
Z.AI China	GLM-5, GLM-4.7, GLM-4.7 Flash (via open.bigmodel.cn)
MiniMax China	MiniMax M2.5 (via api.minimaxi.com)

All providers support both OpenAI-compatible and Anthropic API protocols where applicable. Switch with /provider.

---

New Features

MCP-Powered Agent Tools

When using Z.AI or Z.AI China, the agent now has access to:

• web_search — Search the web for current information
• web_reader — Fetch and parse any URL
• understand_image — Analyze images via MiniMax vision API

/apply Command

Apply file changes from an AI response without re-running the agent:

> /apply
# Finds and applies code blocks from the last AI message

Word Navigation (macOS/Linux)

New keyboard shortcuts for faster input editing:

• Alt+F / Opt+F — Move cursor forward one word
• Alt+B / Opt+B — Move cursor backward one word
• Alt+D / Opt+D — Delete word forward

Rate Limiter

Configurable per-minute rate limits for both API calls and agent commands — prevents runaway usage and API bans. Adjust in /settings.

Session History as Agent Context

The agent now receives recent chat session history as context, enabling better continuity across agent invocations within the same session.

---

Improvements

Performance

• Import resolution cache — smartContext.ts caches resolved import paths in memory, avoiding repeated disk I/O on every agent iteration
• Render optimization — Agent progress stats computed in a single reduce() pass instead of 7 separate .filter() calls
• Async I/O — All file system operations in the renderer now use async APIs

Security

• Shell injection prevention — sanitizeForShell now strips $(), ${}, $var, ;, |, &&, || patterns from user input
• Symlink traversal — validatePath resolves symlinks via realpathSync to prevent path traversal attacks
• Circular symlink detection — listDirectory tracks visited inodes to prevent infinite loops on circular symlinks
• Keychain fallback warning — Explicit warning logged when API keys fall back to plaintext config storage

Type Safety

• Replaced all as any casts in api/index.ts with typed ApiError and TimeoutError classes
• shouldRetry now uses proper instanceof checks

Bug Fixes

• Fixed race condition in agent start — isAgentRunning flag set synchronously before any async work
• JSON.parse calls in project.ts and learning.ts now wrapped in try-catch for corrupt file resilience
• Fixed bracketed paste (Cmd+V) handling in terminal

---

Code Architecture

Large files split for better maintainability:

Before	After
main.ts (2100 lines)	main.ts + commands.ts + agentExecution.ts
tools.ts (1600 lines)	tools.ts + toolParsing.ts + toolExecution.ts + mcpIntegration.ts
agent.ts (1400 lines)	agent.ts + agentChat.ts + agentStream.ts

---

Tests

Test suite expanded from 577 → 721 tests across 23 test files:

New Test File	Coverage
shell.test.ts	validateCommand, executeCommand, execSimple
keychain.test.ts	SmartStorage keychain/fallback/migration
agentChat.test.ts	loadProjectRules, formatChatHistoryForAgent
agentStream.test.ts	SSE stream parsing (OpenAI & Anthropic)
context.test.ts	Session context save/load/merge/summarize
learning.test.ts	Preference learning, custom rules
verify.test.ts	Build/test/typecheck detection
history.test.ts	Undo/redo action history
projectIntelligence.test.ts	Project scan & intelligence caching
api/index.test.ts	chat() retry logic, token tracking

---

Installation

# npm
npm install -g codeep

# curl
curl -fsSL https://raw.githubusercontent.com/VladoIvankovic/Codeep/main/install.sh | bash

# Homebrew
brew upgrade codeep

Full changelog: v1.2.0...v1.2.17

Codeep v1.2.0

What's New

New Provider

• Anthropic — Claude Opus 4.6, Sonnet 4.5, Haiku 4.5 now available as a provider

Features

• Skills execution engine — /commit, /push, /pull, /test, /refactor and 40+ skill commands with multi-step workflows
• Multi-line input — Backslash continuation (\ + Enter) and /multiline toggle mode (Esc to send)
• Token usage tracking — Real-time token counting for both OpenAI and Anthropic streaming, shown in status bar and /status
• Myers diff algorithm — Proper O((N+M)D) diff replacing the greedy heuristic, more precise code previews
• .gitignore awareness — Smart context and file listing now respect .gitignore + 18 built-in ignores
• Expanded search — search_code now covers 30+ file types: Ruby, Kotlin, Swift, PHP, Java, C#, C/C++, Vue, Svelte, YAML, and more
• Git auto-commit — Optional auto-commit after agent tasks (configurable in /settings)
• HTML-to-text — fetch_url now converts HTML to structured text preserving headings, links, code blocks, and lists
• Error localization — Error messages in all 11 supported languages

Improvements

• App.ts split — Extracted Search, Export, Logout panels into separate components
• /protocol now only shows protocols the current provider supports
• /logout correctly filters providers with API keys
• /status shows real message count and token usage
• /help updated with all new commands
• Fixed Enter key handling for terminals sending \r\n
• Fixed logout panel navigation (arrow keys + Enter)
• Fixed skill command output display

Testing

• 335 tests (up from 152) — new test suites for tools, smartContext, agent, and gitignore

Full Changelog

v1.0.0...v1.2.0

Codeep v1.0.0 - Initial Public Release

Features

Multi-Provider Support

• Z.AI (ZhipuAI) - GLM-4.7, GLM-4.7 Flash models
• MiniMax - MiniMax M2.1 model
• Switch between providers with /provider
• Configure different API keys per provider

Project Context Awareness

When started in a project directory, Codeep automatically:

• Detects project type (Node.js, Python, etc.)
• Reads file paths mentioned in your messages
• Attaches file contents to conversations
• Understands your project structure
• Can suggest and apply code changes (with write permission)

Session Management

• Auto-save - Conversations are automatically saved
• Session picker - Choose which session to continue on startup
• Per-project sessions - Sessions stored in .codeep/sessions/
• Rename sessions - Give meaningful names with /rename
• Search history - Find past conversations with /search
• Export - Save to Markdown, JSON, or plain text

Git Integration

• /diff - Review unstaged changes with AI assistance
• /diff --staged - Review staged changes
• /commit - Generate conventional commit messages

Code Block Management

• Automatic syntax highlighting for 12+ languages
• Copy code blocks to clipboard with /copy [n]
• Code blocks are numbered for easy reference

Autonomous Agent Mode

Codeep works as a full AI coding agent that autonomously:

• Creates, edits, and deletes files
• Executes shell commands (npm, git, build, test, etc.)
• Reads and analyzes your codebase
• Loops until the task is complete
• Reports all actions taken

Auto mode (default): Just describe what you want - no special commands needed:

> add error handling to src/api/index.ts
> run tests and fix any failures
> create a new React component for user settings

Manual mode: Use /agent <task> when you want explicit control.