All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- 🖥️ Electron Desktop App: Full rewrite as a native desktop application
- React + Vite + TypeScript frontend with shadcn/ui components
- Integrated PTY terminals (node-pty) for real terminal emulation
- Session persistence for terminal tabs across restarts
- 📋 TaskWizard Enhancements (Auto-Claude Parity):
- Draft Persistence: Form state saved to localStorage on close, restored on open
- Image Paste: Ctrl+V screenshots into description with thumbnail preview
- Model Selector: Choose between Auto, GPT-4o, Claude Sonnet, Gemini Pro
- Enhanced Review: Shows priority, assignee, model, and image count badges
- 💬 Insights Panel: Chat interface for querying codebases with AI
- 📖 Context Viewer: Display
CODEBASE_MAP.mdknowledge map - 🛠️ Surgical File Editing (fs_tool v2):
replace_in_file(): Pattern-based replacement with checksum verificationinsert_into_file(): Line-specific insertionread_file_section(): Read specific line ranges
- 🔀 Git Tool with AI Merge:
get_conflicts(): Parse merge conflict markersresolve_conflicts(): AI-powered conflict resolutioncommit(),push(),create_pr(): Full Git workflow
- Architecture: Shifted from pure Python CLI to Electron + Python backend hybrid
- UI Framework: Dark theme with glassmorphism, micro-animations, premium feel
- 🧪 Comprehensive Test Suite: Full pytest coverage with mocked dependencies
- Unit tests for Brain, Hippocampus, ModelFactory, Delegator, TagParser, Skills
- Integration tests for CLI commands and Wake Protocol
- Shared fixtures for fast, deterministic testing
tests/README.mddocumentation for contributors
- 🔄 GitHub Actions CI/CD: Automated testing on Python 3.10, 3.11, 3.12
- 📊 README Badges: Tests, Coverage, Downloads, Last Commit, Issues
- Resolved git merge conflict markers in
brain.pyandhippocampus.py - Removed duplicate dependencies in
setup.pyandrequirements.txt - Fixed 408 lines of duplicate code in
hippocampus.py
- 🎮 Control Plane Dashboard: Real-time web UI via
squadron server- Live SSE activity streaming
- Command input to send tasks to agents
- Agent status cards with current objectives
- System statistics (active agents, task queue, missions)
- Integrations Panel: Send to Slack, Discord, Jira, GitHub directly from UI
- GitHub PR/Issue creation from dashboard
- Agents Page: Detailed agent profiles with capabilities
- Missions Page: Active/history view with mission trigger
- Console Page: Direct REPL-style chat with agents
- Settings Page: System status and env configuration reference
- 🐝 Swarm 2.0: Enhanced multi-agent orchestration
- LLM-powered intelligent routing (replaces keyword heuristics)
handoff_task()for agent-to-agent delegation with context- Delegation chain tracking
- Task queue with priority support
- 🧬 Evolution Layer: Self-improvement system
skill_registry.pyfor tracking evolved skills- Version control with rollback support
- Quality scoring based on success/failure rates
- Skill creation, validation, and archival
- ⏰ Wake Protocol: Autonomous agent execution
wake_protocol.pyorchestrates trigger → route → execute → reportsquadron wakeCLI command for manual triggering- Automatic ticket reporting after task completion
- 📡 Event Bus: Central pub/sub for real-time activity
- Powers dashboard SSE streaming
- Tracks agent starts, tool calls, completions, errors
- 💬 Agent Communication: Agents can talk to each other via tickets
- Overseer v3.0: Auto-wakes agents when Jira/Linear tickets assigned
- @Mention Support: Tag agents in ticket comments to trigger them
reply_to_ticket()tool for agents to comment on ticketstag_agent()tool to request help from other agents- Autonomous handoffs with context through ticket system
- 🧠 Persistent Memory: Agents remember past work
- Hippocampus v2.0: Agent-specific semantic memory with ChromaDB
save_memory(),recall_memory(),get_memory_context()tools- Conversation and task history tracking
- Memory API endpoints for dashboard integration
- GCP Migration:
- Full deployment to Google Cloud Compute Engine
health_check.pyfor remote diagnostics
- Swarm Overseer: Now uses Gemini for intelligent routing
- Dashboard Frontend: Complete rewrite with shadcn/ui components
- CLI: Added
wakecommand for autonomous execution
(Caleb - 02:45 AM EST)
- Autonomous Quant Skills:
find_strategy_videos: Agent autonomously searches YouTube for new alpha.run_backtest: Agent runs Python backtests validation.get_market_data: Agent checks Alpaca price feeds.
- Intelligent Auditor:
- Smart EOD (3:30 PM): Switched from "Close All" to "Swing vs Cut" decision logic based on PnL/Charts/Time.
- Daily Optimizer (4:30 PM): Auto-triggers research if daily win rate is < 50%.
Note
Notes for Marcus:
- Dashboard UI: The backend now supports a
SWINGstatus for signals. Please ensure the frontendLedgerPageandOpenSignalscomponents correctly display/filter trades withstatus="SWING". They might currently disappear if the query blindly filters forOPEN. - Reporting: The Daily Optimizer posts reports to Slack/Discord. You might want to pipe these into a "Strategy Lab" view on the dashboard eventually.
- MCP Client: Universal bridge for MCP servers.
(Caleb - 01:25 AM EST)
- The Brain: Centralized intelligence router (
brain.py) that decides between conversational replies and tool execution. - Discord Bot: Full
discord.pyintegration with "Neural Link" personas. - Browser Skill:
BrowserToolallows agents (Marcus) to visit websites and capture screenshots. - SSH Skill:
SSHToolallows agents (Caleb) to execute safe remote commands. - File Uploads: Bridge and Listeners now support attaching files generated by tools (e.g., screenshots).
- Masquerading Fix: Resolved "Limited Mode" by ensuring webhooks are used correctly with custom avatars.
- Unified Entry Point:
main.pynow launches Discord Bot, Slack Listener, and Jira Overseer concurrently.
- Planner Skill:
squadron plan --ticket <ID>generates a structuredPLAN.mdfile. - Artifacts: New templates for Implementation Plans.
- Release Fix: Version bump to resolve PyPI collision.
- Local Agent Config:
squadron initnow creates a localsquadron/agents.yaml. - Packaging Fix: Included
agents.yamlandknowledge/*.mdtemplates in PyPI package viaMANIFEST.in.
- Privacy: Default package avatars are now generic Robohash avatars. Users can override them in their local
agents.yaml. - CLI Logic: Prioritizes local configuration files over package defaults for both Knowledge and Agents.
- Documentation: Synced
README.mdon PyPI to reflect v0.2.3 features (Init/Learn/AGPL).
- Init Command (
squadron init): Scaffolds a localsquadron/knowledge/directory and.envfile for new projects. - Local Knowledge Overlay:
squadron asknow prioritizes the localknowledge/folder over the package defaults, allowing per-project customization.
- Librarian Skill (
squadron learn): Auto-generatesCODEBASE_MAP.mdby scanning the repository structure.
- License: Switched from MIT to AGPL-3.0 to enforce open-source reciprocity for network deployments.
- Listener Service (
squadron listen): New command utilizing Slack Socket Mode to allow agents to "hear" and reply to @mentions in real-time. - Dependency: Added
slack_boltfor event handling.
- Dynamic Agent Identities: Agents can now have custom names and avatars (e.g., Marcus, Caleb) supported in Slack and Discord using
assets/andagents.yaml. - Linear Integration: Full support for Linear issues via
--linearflag (create, comment, update status). - RAG-Lite (
squadron ask): New memory module allowing agents to query theknowledge/directory for team context. - Overseer 2.0 (
--exec): The Overseer can now execute arbitrary shell commands when new tickets are detected, enabling "Wake Up" protocols. - Avatars: Hosted assets for Marcus and Caleb.
- CLI Architecture: Refactored to support sub-commands more robustly.
- Documentation: Major updates to
README.mdand intro ofUPDATE.mdfor team onboarding.
- Core CLI: Basic
squadron reportcommand. - Bridges: Initial support for Jira, Slack, Discord, and GitHub.
- Overseer 1.0: Basic polling for new Jira tickets.
- Package: Initial PyPI release structure.