Features
AgileFlow includes powerful features to protect your codebase, automate workflows, and keep your team productive.
Core Features
Skills System
Create, manage, and continuously improve custom AI prompts that learn from your feedback.
- Research-backed creation - Generated from web docs, local documentation, and your codebase
- Cookbook workflows - Multiple use-case specific guides in a single skill
- Self-improving - Skills learn from your corrections and refine automatically
- MCP integration - Connect to Model Context Protocol servers for tool access
- Team shareable - Commit to git and share institutional knowledge
Agent Expertise System
Enable self-improving agents that maintain domain knowledge and accelerate task execution.
- Expertise files - Agents maintain mental models of their domain
- Three-step workflow - Plan (with expertise) → Build → Self-Improve
- Automatic learning - Expertise updates as agents work and make changes
- Validation & metrics - Tools to ensure expertise stays accurate and current
- Focused domains - Specialized knowledge for database, API, UI, testing, CI/CD, and more
Agent Teams
Coordinate multiple domain experts to work together on complex multi-domain tasks using Claude Code's native Agent Teams.
- Dual-mode architecture - Native Agent Teams when enabled, subagent orchestrator as fallback
- Team templates - Predefined teams for fullstack, code-review, builder-validator, and code-logic
- Quality gates - Enforce tests, lint, types, and coverage before work is accepted
- Task synchronization - Bidirectional sync between Claude Code's task list and AgileFlow status
- Auto-detection -
/babysitautomatically suggests teams for multi-domain tasks
Parallel Sessions
Work on multiple features, bugs, or experiments simultaneously with isolated workspaces and boundary protection.
- Git worktrees - Each session gets its own directory and branch
- Boundary protection - Edit/Write hooks prevent accidental cross-session edits
- Easy switching - Use
/add-dirto switch sessions in the same terminal - Session tracking - AgileFlow tracks all sessions with status and activity
- Independent context - Each session maintains its own story and progress
Automation & Execution
Smart Detection
Contextual Feature Router that automatically recommends the right AgileFlow features at the right time.
- 42 feature detectors - Analyze project signals and recommend relevant commands
- Lifecycle-aware - 5 workflow phases ensure recommendations match what you're doing
- Script-based - Deterministic detection, not LLM guesswork
- Anti-repetition - Tracks offered/used/skipped features to avoid nagging
- Configurable - Disable features, adjust thresholds, override priorities
Hooks System
Event-driven automation using Claude Code's official hooks for triggers and responses.
- SessionStart hooks - Run scripts when Claude Code starts
- PreCompact hooks - Preserve context during conversation compression
- PreToolUse hooks - Block dangerous commands before execution
- Ready-to-use scripts - Status line, auto-archival, context preservation
Automations System
Schedule recurring tasks to run automatically without requiring a daemon.
- No daemon required - Runs during user sessions via SessionStart hook
- Multiple schedule types - Daily, weekly, monthly, or custom intervals
- Preset automations - Pre-built automation templates (changelog, CI summary, dependency audit)
- Loop detection - Prevents runaway automation chains
- Timeout protection - Built-in safeguards for stuck processes
Loop Mode
Run Claude autonomously with continuous story execution until epic completion.
- Autonomous story processing - Claude works through an epic without manual intervention
- Test-based verification - Stories verified complete via test suite
- Configurable limits - Set max iterations and timeouts
- Smart detection - Auto-enabled for epics with clear acceptance criteria
- Multi-mode support - Normal (test-based), Visual (UI), or Coverage-focused modes
Visual Mode
Screenshot verification system for UI development with automated visual progress tracking.
- Verified-prefix protocol - Screenshots must be reviewed and renamed with
verified-prefix - Integrated with Loop Mode - Adds visual verification step in autonomous execution
- Multiple states capture - Test default, hover, disabled, and other UI states
- Auto-cleanup - Optional cleanup of verified screenshots after completion
Compact Context Preservation
Preserve critical agent knowledge and state during automatic context compaction.
- Context-aware compaction - Critical rules preserved when Claude Code compresses context
- Priority-based preservation - Critical agents always preserved
- State continuity - Current story, blockers, and progress maintained
- Long conversation support - Sustained agent focus over days of work
- Zero-overhead - Automatic, no manual configuration needed
Decision Making & Analysis
AI Council
Get balanced strategic perspectives on complex architectural and business decisions.
- Three perspectives - Optimist (opportunities), Advocate (risks), Analyst (trade-offs)
- Parallel deliberation - All three agents analyze simultaneously
- Debate mode - Multi-round deliberation where agents respond to each other
- Consensus synthesis - Final recommendation balancing all viewpoints
Logic Audit
Find logic bugs with multi-agent specialized analyzers and consensus voting.
- Five analyzer types - Edge cases, invariants, control flow, type errors, race conditions
- Parallel analysis - All analyzers run simultaneously for efficiency
- Confidence scoring - CONFIRMED (2+ agree), LIKELY (evidence-based), INVESTIGATE
- Actionable reports - Prioritized findings with reproduction steps
RLM (Recursive Language Models)
Analyze complex large documents without context bloat using virtualization.
- Document virtualization - Never loads full document into context
- Programmatic search - Use document-repl for targeted extraction
- Multi-hop reasoning - Support for cross-referenced content (contracts, specs)
- Iteration limits - Built-in budget constraints and iteration tracking
Research Pipeline
Structured research workflow for gathering and synthesizing knowledge.
- Research prompts - Generate detailed 200+ line prompts for external AI tools
- Import & organize - Save research findings from external sources
- Analysis - Extract structured insights from research notes
- Synthesis - Combine multiple research sources into actionable conclusions
API Server
REST API for external integrations and dashboard applications.
- Read-only access - Exposes state without mutations
- Localhost-only - Secure by default (127.0.0.1)
- JSON endpoints - Access sessions, status, tasks, and metrics
- No daemon needed - Starts and stops on demand
Setup & Safety
Configuration System
Unified system for managing hooks, safety features, and automation with preset profiles and individual feature control.
- Git setup - Initialize repository and remote connections
- Hooks system - Event-driven automation (SessionStart, PreCompact, Stop)
- Status line - Custom status bar in Claude Code
- Archival - Auto-manage status.json file size
- CI/CD - Automated testing with GitHub Actions, GitLab, CircleCI
- Attribution - Control AI disclosure in git commits
- And more - PreCompact context, verification, token management
IDE Integrations
Install and configure AgileFlow for your favorite IDE with automatic command, agent, and skill installation.
- Multiple IDEs - Claude Code, Cursor, Windsurf, OpenAI Codex CLI
- Auto-detection - Detect which IDEs are configured in your project
- Damage Control - Optional safety system for Claude Code
- Dynamic content - Commands and agents automatically injected on install
- Multi-IDE support - Use the same project across multiple IDEs
Damage Control
Protect your codebase from destructive commands using PreToolUse hooks.
- Bash command validation - Block dangerous patterns like
rm -rf /,git push --force - File access control - Protect secrets, keys, and critical files
- Customizable rules - Add project-specific protection patterns
- Ask-first mode - Require confirmation for risky operations
- Three protection layers - Bash, Edit, and Write tool validation
Quick Reference
| Feature | Purpose | Get Started |
|---|---|---|
| Skills | Custom AI prompts that learn | /agileflow:skill:create |
| Agent Expertise | Self-improving domain knowledge | Built-in |
| Agent Teams | Multi-domain expert coordination | /agileflow:team:start |
| Parallel Sessions | Work on multiple features | /agileflow:session:new |
| Smart Detection | Contextual feature recommendations | Built-in with /agileflow:babysit |
| Hooks System | Event-driven automation | /agileflow:configure |
| Automations | Scheduled recurring tasks | /agileflow:automate |
| Loop Mode | Autonomous epic execution | /agileflow:babysit EPIC=... MODE=loop |
| Visual Mode | UI screenshot verification | /agileflow:babysit EPIC=... VISUAL=true |
| AI Council | Strategic decision perspectives | /agileflow:council <question> |
| Code Logic Audit | Find logic bugs with consensus | /agileflow:code:logic [file] |
| RLM | Large document analysis | /agileflow:rlm DOCUMENT=... |
| Research Pipeline | Structured research workflow | /agileflow:research:ask |
| Compact Context | Preserve state through compression | Built-in |
| Configuration | Setup and customize workflow | /agileflow:configure |
| IDE Integrations | Multi-IDE support | npx agileflow setup |
| Tmux Integration | Multi-window terminal sessions | af |
| Damage Control | Block dangerous commands | /agileflow:configure |
| API Server | REST API for external integrations | /agileflow:api ACTION=start |
Next Steps
- Start with Setup - Run
/agileflow:configureto set up your project - Enable Safety - Configure Damage Control to protect your codebase
- Learn Skills - Create custom prompts for your workflow
- Try Parallel Sessions - Work on multiple features simultaneously
On This Page
FeaturesCore FeaturesSkills SystemAgent Expertise SystemAgent TeamsParallel SessionsAutomation & ExecutionSmart DetectionHooks SystemAutomations SystemLoop ModeVisual ModeCompact Context PreservationDecision Making & AnalysisAI CouncilLogic AuditRLM (Recursive Language Models)Research PipelineAPI ServerSetup & SafetyConfiguration SystemIDE IntegrationsDamage ControlQuick ReferenceNext Steps