Skip to content

CHANGELOG.md

Latest commit

 

History

History
331 lines (235 loc) · 12.1 KB

CHANGELOG.md

File metadata and controls

331 lines (235 loc) · 12.1 KB

Changelog

All notable changes to this project will be documented in this file.

[1.4.2] - 2025-01-05

Added

  • Explicit model selection for all Task() calls in orchestration skill
    • Every Task() example now includes model= parameter
    • Model tier framework: haiku (errand runner), sonnet (capable worker), opus (critical thinker)
    • Guidance on when to use each model based on task type

Changed

  • All commands now use npx cc-mirror for portability - no global install required

    • Updated task-manager skill with npx prefix
    • Updated README.md command examples
    • Updated team-mode.md CLI examples
    • Updated CLI help text (help.ts)
    • Updated AGENTS.md and architecture docs
    • Updated issue templates

  • Model Selection section added to SKILL.md with tier descriptions

  • Updated 12 skill reference files with explicit model parameters:

    • references/tools.md - Model selection table and examples
    • references/patterns.md - All pattern examples with models
    • references/examples.md - All workflow examples with models
    • 8 domain guides with appropriate models per task type

Model Selection Guidelines

Model Best For
haiku Exploration, searching, fetching context, data discovery - spawn many
sonnet Implementation, test generation, documentation, structured work
opus Code review, security analysis, architecture, conflict resolution

[1.1.5] - 2025-01-05

Added

  • Role Detection in orchestration skill

    • Skill now detects if it's the main orchestrator or a spawned worker
    • Workers skip orchestration and execute tasks directly
    • Prevents recursive orchestration chaos

  • Worker Agent Prompt Template

    • Required WORKER preamble for all spawned agents
    • Clear rules: execute task, use tools directly, no sub-agents, no task management
    • Example prompts included in SKILL.md and references/tools.md

  • Tool Ownership section

    • Clear separation: orchestrator tools vs worker tools
    • Orchestrator: TaskCreate, TaskUpdate, TaskGet, TaskList, AskUserQuestion, Task
    • Workers: Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, LSP

  • Complete task lifecycle in "What you DO"

    • Expanded from 5 to 8 steps
    • Added: Set dependencies, Find ready work (TaskList), Mark complete (TaskUpdate resolved)

Changed

  • Flow diagram now shows complete lifecycle

    • Added "Find Ready Work" step with TaskList
    • Added "Mark Complete" step with TaskUpdate(resolved)
    • Shows loop back to find more ready work

  • Agent scaling is now guidance-based, not quota-based

    • Replaced "3 agents minimum is non-negotiable"
    • New table: Quick lookup (1-2), Multi-faceted (2-3), Full feature (4+ swarm)
    • "Match the swarm to the challenge"

  • Subagent Prompting Guide in references/tools.md

    • WORKER preamble now required first element
    • "Four Elements" → "Five Elements" (added preamble)
    • All examples updated with preamble

Fixed

  • Agents no longer try to re-orchestrate when spawned
  • Task lifecycle now explicitly includes resolution step
  • Clearer separation between orchestrator and worker responsibilities

[1.1.4] - 2025-01-05

Changed

  • Renamed skill: multi-agent-orchestrator → orchestration

    • Shorter, cleaner name
    • Updated all references in code, tests, and documentation

  • Completely rewritten orchestration skill with enhanced personality and capabilities

    • New "Conductor on the trading floor" identity with warmth and swagger
    • Dynamic personality that adapts to user energy (excited, frustrated, curious, etc.)
    • Strict "Iron Law" enforcement: orchestrator NEVER uses Read/Write/Edit/Bash directly
    • Minimum 3 agents per request, swarm everything philosophy
    • Rich milestone celebrations and progress updates
    • Signature branding (─── ◈ Orchestrating ──)

  • Maximal AskUserQuestion guidance

    • Always use 4 questions (the max) when gathering context
    • Always use 4 options per question (the max)
    • Rich descriptions with no length limit - explain trade-offs, implications, examples
    • "Be a consultant, not a waiter" philosophy

Added

  • Auto-approve orchestration skill: Skill(orchestration) added to permissions.allow when team mode enabled

    • No more permission prompts to load the skill

  • Team pack system prompt: system-prompt-orchestration-skill.md

    • Instructs loading orchestration skill before any response

  • Skill tool override: tool-description-skill.md

    • CRITICAL instruction to load orchestration skill first in every conversation

  • Domain expertise routing preserved with rich reference files:

    • references/patterns.md - Fan-Out, Pipeline, Map-Reduce, Speculative patterns
    • references/tools.md - Enhanced tool usage with maximal questioning philosophy
    • references/examples.md - Complete workflow examples
    • references/guide.md - User-facing documentation
    • 8 domain-specific guides (software-development, code-review, testing, etc.)

Fixed

  • Orchestrator now consistently delegates to agents instead of doing work directly
  • AskUserQuestion examples now show comprehensive 4-question, 4-option patterns

[1.1.3] - 2025-01-04

Fixed

  • useEffect infinite loop when toggling team mode on existing variants

    • Added ref guards to all async TUI hooks to prevent concurrent execution
    • Stabilized refreshVariants callback with useCallback
    • Hooks affected: useTeamModeToggle, useVariantUpdate, useVariantCreate, useModelConfig, useUpdateAll

  • Team mode visibility in configuration/summary screens

    • TUI SummaryScreen now shows team mode status before variant creation
    • CLI printSummary now shows team mode in all modes (quick, interactive, non-interactive)
    • Provider-specific prompt pack routing info (zai-cli vs MCP routing)

  • Skill tool examples mismatch - Added skill clarification spec to prevent confusion

    • Prompt pack now clarifies that skill examples (commit, review-pr, pdf) are illustrative only
    • Directs users to check <available_skills> for actually installed skills

Added

  • Team Pack prompt files for enhanced team mode guidance

    • task-management-note.md - Clarifies TodoWrite deprecation vs Task* tools in team mode
    • tasklist.md, taskupdate.md, task-extra-notes.md - Enhanced Task* tool descriptions

  • TeamModeScreen TUI component for team mode selection during variant creation

  • Comprehensive tests

    • test/e2e/blocked-tools.test.ts - Tests for blocked tools configuration
    • test/e2e/team-mode.test.ts - E2E tests for team mode enable/disable/toggle
    • test/provider-matrix.test.ts - Provider feature matrix validation
    • test/tui/TeamModeScreen.test.ts - TeamModeScreen component tests
    • test/tui/ModelConfigScreen.test.ts - ModelConfigScreen component tests

Changed

  • Removed deprecated promptPackMode parameter (maximal mode removed, minimal is now default)
  • Shell env option now only shown for zai provider in summary screens
  • Team mode description now includes details: "on (orchestrator skill, TodoWrite blocked)"

[1.1.2] - 2025-01-04

Fixed

  • Suppress verbose tweakcc output during CLI variant creation

[1.1.1] - 2025-01-04

Fixed

  • Mirror and CCRouter providers no longer prompt for API key (they use OAuth or optional keys)
  • Cleaned up CLI output formatting

Removed

  • Removed Twitter/X share URL from create output

[1.1.0] - 2025-01-04

Added

  • Team Mode - Multi-agent collaboration with shared task management

    • --enable-team-mode flag for create and update commands
    • Toggle team mode on/off in TUI variant management screen
    • Patches Claude Code CLI to enable TaskCreate, TaskGet, TaskUpdate, TaskList tools
    • Automatic backup of CLI before patching (cli.js.backup)

  • Mirror Claude Provider - Pure Claude Code variant with enhanced features

    • No proxy - connects directly to Anthropic's API
    • Team mode enabled by default
    • Silver/chrome theme with electric blue accents
    • OAuth or API key authentication (standard Claude Code auth flow)

  • Multi-Agent Orchestrator Skill - Automatically installed when team mode is enabled

    • "The Conductor" identity for elegant multi-agent orchestration
    • AskUserQuestion as mandatory tool (never text menus)
    • Background agents by default (run_in_background=True)
    • 8 domain-specific reference guides (code review, testing, devops, documentation, etc.)
    • Managed skill marker (.cc-mirror-managed) for safe updates without overwriting user customizations

  • Documentation

    • docs/features/team-mode.md - Complete team mode guide with architecture diagrams
    • docs/features/mirror-claude.md - Mirror Claude provider documentation
    • docs/architecture/overview.md - System architecture overview
    • Updated AGENTS.md with team mode and orchestrator skill sections

Changed

  • Bundle script now copies skills to dist/skills for npm distribution
  • Enhanced TUI with team mode toggle in variant actions screen
  • Provider selection includes Mirror Claude with education content

[1.0.4] - 2025-01-04

Changed

  • Removed broken ASCII art success banner from completion screen
  • Streamlined Z.ai prompt pack (removed verbose setup/advanced sections)
  • Simplified MiniMax prompt pack (removed redundant auth section)

[1.0.3] - 2025-01-03

Changed

  • Removed 5 unused dependencies: gradient-string, ink-big-text, ink-box, ink-gradient, ink-spinner
  • Production dependencies reduced from 10 to 5
  • Package tarball size reduced to 88.5 kB

Fixed

  • Fixed bin path to use relative path (./dist/cc-mirror.mjs)
  • Added missing @eslint/js dev dependency

[1.0.2] - 2025-01-03

Changed

  • Upgraded to Ink 6.6.0 and React 19
  • Updated all ink-* packages to latest versions
  • Fresh dependency tree with improved compatibility

[1.0.1] - 2025-01-03

Fixed

  • Fixed npx compatibility by keeping React/Ink as external dependencies
  • Resolved dynamic require and ESM bundling issues
  • Bundle now properly delegates React ecosystem to npm

[1.0.0] - 2025-01-03

Added

  • First public release
  • Claude Code Router support (route to local LLMs via CCRouter)
  • RouterUrlScreen for simplified CCRouter configuration
  • Provider intro screens with setup guidance and feature highlights
  • Feedback screen with GitHub repository links
  • Beautiful README with screenshots and n-skills style formatting

Changed

  • Removed LiteLLM provider (replaced by Claude Code Router)
  • CCRouter no longer requires model mapping (handled by CCRouter config)
  • Simplified provider selection flow with better education
  • Updated provider content to emphasize local LLM support
  • Version bump to 1.0.0 for first stable release

Fixed

  • CCRouter flow no longer shows "model mapping incomplete" warning
  • Settings-only updates preserve binary patches (fixes theme reset issue)
  • All linting errors resolved
  • React hook dependency warnings fixed

[0.3.0] - 2025-01-02

Added

  • Colored ASCII art splash screens for each provider
    • Z.ai: Gold/amber gradient
    • MiniMax: Coral/red/orange gradient (matching brand)
    • OpenRouter: Teal/cyan gradient
    • LiteLLM: Sky blue gradient
  • Async operations for live TUI progress updates
  • MIT License

Changed

  • Renamed "Local LLMs" provider to "LiteLLM" throughout
  • Footer layout: creator info on left, social links stacked on right
  • tweakcc option now shows CLI command (avoids TUI-in-TUI conflict)
  • Prepared package.json for npm publish (removed private flag, added metadata)

Fixed

  • Progress bar and step animations now update in real-time
  • MiniMax colors now match official brand (coral/red, not purple)

[0.2.0] - 2025-01-01

Added

  • Full-screen TUI wizard
  • Brand theme presets (zai, minimax, openrouter, local)
  • Prompt packs for enhanced system prompts
  • dev-browser skill auto-installation
  • Shell environment integration for Z.ai

Changed

  • Restructured to use ink for TUI
  • Modular provider templates

[0.1.0] - 2024-12-30

Added

  • Initial release
  • CLI for creating Claude Code variants
  • Support for Z.ai, MiniMax, OpenRouter, Local LLMs
  • tweakcc integration for themes
  • Variant isolation with separate config directories