Easy Code v1.1.27

Merge branch 'ls-dev' into 'master'

fix: delegate_to_agent 工具注册探测 + lark-cli 默认排除敏感权限 + 品牌升级清理

See merge request ai_native/DeepVCode/DeepVcodeClient!532

Easy Code v1.1.16

New Features

Default Resume Last Session

When launching the Easy Code CLI, you no longer need to manually add the -c parameter. By default, it automatically restores the last conversation session in the current directory, continuing seamlessly with the existing context.
To start a brand-new session, add --no-continue.

/goal — Goal-Driven Mode (Independent Reviewer)

In /goal mode, an independent review mechanism is introduced. Instead of having the current model decide whether the task is complete, a separate supervisor model (deepseek-v4-flash) objectively verifies completion against the “pass criteria”/success characteristics.
If the review passes, the goal is officially marked as completed. Otherwise, it provides detailed reasons for what’s not yet met and continues to move forward.

/loop — Watchdog Command

A new /loop command is added to start a session-level watchdog in the current conversation, continuously monitoring task execution status.
Supports --expires to customize the watchdog lifetime (default: 30 minutes).
The VSCode plugin UI has also been updated to display watchdog status.

Stability Improvements

SubAgent Timeout & Memory Protection

• Per-round timeout: If a single streaming response exceeds 5 minutes, it will be automatically interrupted and return the partially collected results to prevent infinite waiting.

• Tool completion timeout: Reduced from 10 minutes to 3 minutes to reduce the perceived “stuck” time.

• Overall hard timeout: A 30-minute hard limit is added at the TaskTool layer. When it times out, it is force-canceled and returns partial results.

• Memory protection: executionLog is automatically truncated when it exceeds 200 entries. If conversation history compression fails, a hard cutoff fallback is enabled (keeps the system instructions + the last 25 rounds) to prevent OOM.

• Resource leak fixes: The tool completion timeout timer now properly clears with normal completion, preventing timer leaks.

Internationalization

SubAgent UI Text i18n

All SubAgent status texts in both the CLI and the VSCode plugin (e.g., “thinking and executing”, “tool calling”, “execution time”, “token usage”, “rounds”, etc.) have been fully internationalized. Switching between Chinese and English is now smooth.

Bug Fixes

• Fixed a bug in /goal mode where the supervisor model’s feedback could incorrectly overwrite the main model’s systemInstruction. This is resolved by correctly isolating systemInstruction in runForkedAgent.
• Fixed a TypeScript syntax error in the VSCode plugin test files’ Lexical mock class property declarations (: → =).

Build Optimization

In the core package, tsconfig.json now excludes src/**/*.test.ts. Production builds are no longer affected by test file types, significantly improving build speed.

Easy Code v1.1.12

Merge branch 'ls-dev' into 'master'

Release v1.1.12: Agent delegation, /btw, mid-turn injection, hot-restart fixes

See merge request ai_native/DeepVCode/DeepVcodeClient!498

Easy Code v1.1.7

Merge branch 'ls-dev' into 'master'

fix(feishu): parse card 2.0 user_dsl, table component, and markdown-type cells

See merge request ai_native/DeepVCode/DeepVcodeClient!492

Easy Code v1.1.6

New: Feishu merge-forward message parsing - the AI can now read forwarded chat records (sender, timestamp, content) to summarize discussions, draft replies, and extract key info.

v1.1.4

Easy Code v1.1.4 Released 🎉

New Features

1. Send /feishu restart from Feishu
   Reminds/restarts the remote CLI process and automatically restores the Feishu gateway — takes only 3 seconds. Any exception can be recovered easily.

2. Improved /feishu status display
   Now uses tables to clearly show the binding relationships between projects and chat groups.

3. Natural language can trigger restart
   e.g. “Please restart yourself”, and EasyCode will restart accordingly.

4. Support /goal from Feishu channels
   Send /goal in a channel, then fill out the form to get started immediately.

Fixes

• Fixed the bug where after /stop, sending a message would remain queued forever.
• Fixed: In a Feishu channel, the ask_user_question tool could fail most of the time.

EasyCode v1.1.3

Changed

• Rebrand: DeepV Code → Easy Code
• Renamed the package to easycode-ai and introduced the new easycode launch command.
• The legacy dvcode command continues to work and now prompts users to switch to easycode.
• Unified the brand across CLI, VSCode, and Feishu so the experience is consistent on every surface.

Added

• Automatic Migration of Legacy Data
  • On upgrade, legacy user and project data is migrated automatically into ~/.easycode-user and ./.easycode, so existing users can switch over seamlessly with dvcode -u.
• Feishu Auto-Resume Conversations
  • Feishu mode now supports automatic conversation resume — even after a restart, you can pick up right where you left off without any extra repetitive steps.

Fixed

• Open-Source Model HTTP 400 Fix
  • Fixed potential HTTP 400 errors that could occur with certain open-source models.

v1.0.400

DeepV Code v1.0.400

Gradually sync the commercial release into the open-source version

• High-frequency Feishu updates: upgraded the Feishu office assistant and collaboration board, improved compatibility for Feishu voice message handling, refined interaction card supplementary input, enhanced background task management, and implemented tool lifecycle governance.
• LLM Wiki proactive consumption: the system now proactively reads existing project architecture/knowledge from .llm-wiki/ before exploring the codebase.
• Terminal & release stability: improved terminal visual noise reduction for Feishu tool calls/history; enhanced release pipeline stability with stronger pre-checks and double-build validation.
• Background services & tasks: added automatic detection and background running for long-lived service commands (returns a Task ID); added tools to list/inspect/terminate background tasks safely.
• Task/goal & tooling governance:
  • /goal (goal-driven mode) behavior refined: goal_achieved is dynamically loaded only in goal mode and unloaded afterward.
  • read_lints registered only when a connected VS Code environment is detected; hidden in standalone terminals to reduce prompt size and mis-invocations.
• Feishu interaction UX improvements:
  • Card state transitions skip throttling for more immediate UI.
  • Feishu card outputs show diff-style summaries instead of full JSON dumps.
  • Added a “I have other ideas” entry to continue input in normal chat flow more naturally.
• Memory/OOM protection: trimmed in-memory prompts and cleared recentToolCalls after workflow end to reduce long-session OOM risk.
• Safer transpilation: strengthened ES→CJS transpiling by using a lexical state machine that skips strings/comments to prevent incorrect transforms of export default.
• Model pool hygiene: filtered out TTS-marked speech models to prevent non-dialogue models from polluting the backup pool.
• Robustness & fixes:
  • Win32 path validation improved (case-insensitive checks).
  • Improved type safety for InspectScript options via ToolConfirmationOutcome.
  • Fixed Feishu mode stability issues and multiple tool-related edge cases (e.g., read_file compact line parsing, media/image compatibility, duplicate message filtering, reconnection reliability).
• Productivity features:
  • Added Order Invoice (billing) generation & A4-friendly printable HTML with auto “Bill To” from user email and editable fields.
  • Added /todo manual task management and automatic cleanup after goal_achieved.
  • Extended default shell command timeout (to 5 minutes) for long tasks.
• Feishu Status Dashboard: introduced FeishuStatusDashboard to show robot name, connection status, bound projects, and real-time logs; improved streaming rendering and reduced prompt exposure for privacy.
• Additional enhancements: improved context compression reliability, AskUserQuestion tolerance, CLI/VSCode model selector reliability, dedup/reconnect handling, and UI consistency (reduced visual interference, unified feedback card titles).**

DeepV Code v1.0.306

Changelog

Improved

• Optimized skill loading compatibility.

Updated

• Simplified startup UI for a cleaner appearance.

Fixed

• Resolved compatibility issues with certain models.

DeepV Code v1.0.302

1. Memory system optimization: Removed the logic for reading memory from gemini.md and claude.md, further simplifying the memory system’s storage structure。
2. Healthy use reminder optimization: Reduced the countdown reminder time for healthyUse from 300 seconds to 60 seconds, making it easier to skip。
3. Prompt simplification: Removed unnecessary prompt messages related to the /bug and /help commands, making the interface cleaner。