13 Proven Missions, One /coord Command
From emergency fixes to complete MVPs, execute complex development missions with project-local agents that understand your codebase.
BUILD
New feature development with full stack implementation
Key Features
- Full-stack feature implementation
- Database schema design
- API endpoint creation
- +3 more features
FIX
Emergency bug resolution and system recovery
Key Features
- Root cause analysis
- Critical bug fixes
- System stability restoration
- +3 more features
MVP
Complete minimum viable product development
Key Features
- Full product architecture
- Core feature implementation
- User interface design
- +3 more features
DEPLOY
Production deployment and infrastructure setup
Key Features
- Production environment setup
- CI/CD pipeline configuration
- Domain and SSL setup
- +3 more features
OPTIMIZE
Performance optimization and system tuning
Key Features
- Performance profiling
- Database query optimization
- Caching strategy implementation
- +3 more features
SECURITY
Security audit and vulnerability remediation
Key Features
- Security vulnerability scanning
- Penetration testing
- Authentication hardening
- +3 more features
REFACTOR
Code quality improvement and technical debt reduction
Key Features
- Code structure improvement
- Design pattern implementation
- Dependency updates
- +3 more features
DOCUMENT
Comprehensive technical documentation creation
Key Features
- API documentation
- Architecture guides
- Setup instructions
- +3 more features
MIGRATE
System migration and technology stack upgrades
Key Features
- Data migration
- Zero-downtime deployment
- Rollback procedures
- +3 more features
INTEGRATE
Third-party service and API integration
Key Features
- API integration
- Webhook handling
- Error handling
- +3 more features
RELEASE
Production release management and versioning
Key Features
- Version management
- Release notes
- Changelog generation
- +3 more features
DEV-SETUP
Development environment initialization for new projects
Key Features
- Repository setup
- Architecture documentation
- Development tooling
- +3 more features
DEV-ALIGNMENT
Analyze and document existing codebase for team alignment
Key Features
- Codebase analysis
- Architecture review
- Technical debt assessment
- +3 more features
Agents that cannot game their own tests
The trust problem with autonomous agents is simple: an agent that can edit the test can pass the test. v6.2 closes that gap. Loops are watched first and merged by you. A passing gate means the work was done, not that the bar was lowered.
Read-only quality gates
The files that judge an agent's work are off limits to the agent. A passing gate means the work was done, not that the bar was lowered.
.quality-gates.jsonandgates/locked by permissions.deny- A Bash-write guard hook backs up the permission block
- No agent can rewrite its own success criteria
Default-fail verification
Every success criterion starts failing. It flips to pass only on captured command output. Claiming done is not enough: the agent has to show the run.
- Criteria fail by default until proven
- Pass requires real, captured output
- No roleplayed completion
Ratchet mission-optimize
Optimisation runs as a measured loop in an isolated git worktree. It keeps a change only if it beats a baseline and hard-reverts the rest. It never auto-merges, so you stay the judge.
- Beat the baseline or get reverted
- Every attempt logged, the loop caps itself
- You decide at merge time
code-review-loop skill
A read-only critic raises evidence-backed findings. A separate read-write fixer addresses only those findings. The pair re-audits until the work is clean or the loop caps out.
- Critic cannot edit the code it judges
- Fixer addresses only raised findings
- Re-audit until clean or capped
Phase-gated meta-loop
/coord continue advances only on evidence and converges on two clean rounds. It spends a per-phase error budget, then escalates instead of grinding.
- Converges on two clean rounds
- Per-phase error budget, then escalates
- Restarts from the last passed gate
Phase 1 & 2 Modernization Features
Enterprise-grade capabilities for zero context loss and professional delivery
Context Preservation System
Zero context loss across multi-agent workflows. v6 consolidated tracking from 5 files to 3 for leaner, faster handoffs.
- agent-context.md as the single accumulator for mission state
- Phase Handoff blocks (Findings / Decisions / Warnings / Open Items / Evidence)
- evidence-repository.md for artifacts
Extended Thinking
Three thinking modes for optimal decision-making: standard, think hard, and think harder for critical components.
- Standard mode for routine tasks
- Think hard for complex features
- Think harder for security-critical code
Field Manual
Professional-grade architecture documentation with comprehensive system design guides and best practices.
- Architecture SOP templates
- Security-first development principles
- Production-ready templates
Memory Management
Agents learn and remember project patterns, technical decisions, and optimization insights across sessions.
- Technical patterns storage
- Decision history tracking
- Lessons learned repository
Security-First Architecture
Critical software development principles enforced at every level. Never compromise security for convenience.
- Root cause analysis mandatory
- Security feature validation
- Strategic solution checklist
Tool Permission Framework
Granular tool permissions for each specialist agent, ensuring security and proper access control.
- Role-based tool access
- MCP integration support
- Fallback strategies documented
Karpathy Constitution
Seven principles every specialist applies by default. Replaces ~150 lines of "always do X" prompt rules with a single posture that actually changes how agents work.
- PAUSE-AND-PLAN before any non-trivial change
- State assumptions out loud, prefer minimal diffs
- Verify by running. Refuse to roleplay completion
Quality-Gate Hooks
v6 ships .claude/settings.json with native Claude Code hooks. Quality checks that used to be prompt suggestions are now real enforcement.
- PostToolUse: tsc / ruff / rubocop on every Edit/Write
- PreToolUse: confirmation for destructive Bash (rm -rf, force push, hard reset)
- Auto-skip when toolchain is absent. Promote advisory to blocking with one flag
3-Tier Skills Model
Skills are the unit of domain knowledge in v6. Aligned with Anthropic's open Agent Skills standard so they're portable across the ecosystem.
- Tier 1: behavioural (loaded via CLAUDE.md)
- Tier 2: project-domain (your own skills/ folder)
- Tier 3: marketplace (7 SaaS skills shipped: auth, payments, multitenancy, billing, email, onboarding, analytics)
6 Slash Commands for Workflow Automation
Pre-configured workflows for common development scenarios
/coordMission Orchestration
Orchestrate multi-agent missions with systematic workflows. Execute build, fix, mvp, and more.
/reconUI/UX Reconnaissance
Execute comprehensive UI/UX assessment using RECON Protocol for design audits.
/design-reviewDesign Review
Complete comprehensive design review of pending changes with systematic assessment.
/pmdPost Mortem Dump
Analyze failures and suggest improvements with root cause analysis protocol.
/reportProgress Reports
Generate progress reports for stakeholders with metrics and accomplishments.
/meetingMeeting Facilitation
Conversational meetings with AGENT-11 specialists for strategic discussions.
11 Specialist Agents with Proven Capabilities
Each agent has specific tool permissions and coordination protocols
THE COORDINATOR
Mission orchestration and agent coordination. Manages task delegation and workflow execution.
THE DEVELOPER
Full-stack development and implementation. Ships clean, working code fast with tests.
THE ARCHITECT
System design and technical architecture. Plans scalable solutions and technology stack.
THE TESTER
Quality assurance and automated testing. Validates functionality and edge cases.
THE OPERATOR
DevOps, deployment, and infrastructure. Manages production environments and CI/CD.
THE STRATEGIST
Product requirements and strategic planning. Defines vision and execution roadmap.
THE DESIGNER
UI/UX design and user experience. Creates intuitive interfaces and design systems.
THE DOCUMENTER
Technical documentation and knowledge base. Creates comprehensive guides and references.
THE ANALYST
Data analysis and performance metrics. Provides insights and optimization recommendations.
THE MARKETER
Growth strategy and user acquisition. Plans launches and marketing campaigns.
THE SUPPORT
Customer support and issue resolution. Handles user inquiries and feedback loops.
MCP Integration - Native Tool Search
AGENT-11 uses Claude Code's native tool-search primitive. Specialists discover MCP servers at runtime instead of loading a fixed list into every session, cutting context overhead and supporting long-running autonomous operation.
Native Tool Deferring (new in v6)
v6 retired the hand-rolled MCP profile system in favour of Claude Code's native primitive: ENABLE_TOOL_SEARCH=auto. Specialists discover MCP tools on demand instead of loading every tool into every session.
tool_search_tool_regex only when a task needs them./mcp-switch command. The framework gets out of the way..claude/settings.json ships with the flag enabled. No extra setup. Works on every fresh install.Tools load only when a task needs them, so the context budget goes back to the user's actual problem.
Advanced Capabilities
Enterprise-grade features for maximum efficiency and performance
Memory Management
Persistent file-based memory system that learns and remembers across sessions.
- Cross-session learning
Knowledge persists between restarts
- Structured XML storage
Clear, human-readable memory files
- Long-running autonomous operation
Memory keeps work going across sessions
Extended Thinking
Strategic reasoning modes for complex tasks requiring deep analysis.
- 3 thinking modes
Standard, think hard, and think harder
- Strategic allocation
Use deeper thinking where it matters most
- Cost-optimized
Match thinking depth to task complexity
Context Optimization
Intelligent context management for optimal performance.
- Native Tool Search
Tools load via defer_loading only when needed
- Strategic context clearing
Preserve critical information
- Memory integration
Long-term knowledge persistence
- Multi-session workflows
Seamless pause and resume
Ready to Execute Your First Mission?
Deploy project-local agents that understand your codebase and execute missions with precision.