- restore: recovered lost
create-issue.mdslash command from commit 8175b4c - command:
/fb:create-issuenow available for generating comprehensive issue documentation from working plan insights
- statusline: read/write
.claude/statusline/state.jsonat the git repository root (gitRoot) instead of the current working directory - init: always overwrite installed statusline templates during refresh to ensure the latest monitor is deployed
- init: install to the git repository root even when run from a nested subdirectory (avoids creating nested
.claude/trees)
- docs: update instructions and version references to v2.4.0
- fix: write SessionStart hooks to project-level .claude/settings.json (reset statusline cache on session start)
All notable changes to Flashbacker will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
- Statusline improvements: Robust model detection (Sonnet/Opus/Claude 4 and [1M] hints)
- Stateful output: Cache last good line to avoid zero flicker
- JSONL fallback: Use latest
~/.claude/projects/-<project>/*.jsonlwhentranscript_pathmissing - Session-aware cache: Reuse only within same session; derive
session_idwhen available - Hooks: SessionStart clears
.claude/statusline/state.jsonon new/resumed sessions
flashback init --statusline-registerregisters project-level statusline; removed global flag
- Claude Context Status Line Monitor: New script
claude_context_monitor.jswith real token usage and color-coded pressure- Installed via
flashback initto.claude/statusline/ - Detects 1M vs 200K context windows based on Claude model
- Example transcripts and README included under
templates/.claude/statusline/
- Installed via
- Init process now installs statusline templates and marks scripts executable
- NEW TASK TEMPLATES: Added comprehensive task management system with 4 new slash commands
/fb:create-tasks: Generate atomic, AI-friendly task breakdown from comprehensive issues/fb:update-tasks: Mark tasks complete, update progress tracking, maintain dashboards/fb:tasks-status: Display comprehensive progress dashboard across all active issues/fb:work-task: Focus AI attention on single atomic task with full context and clear deliverables
- Enhanced project management capabilities with granular task tracking and atomic deliverable focus
- Surgical Discipline Integration: All task templates now enforce fix-master principles
- Anti-placeholder enforcement: Strong language against creating empty/TODO files
- Focused implementation: Prevents E2E test creation for incomplete features
- Atomic task discipline: One task, one deliverable, no scope creep
- Quality standards: Working code required, manual testing first, targeted tests only
- Created 4 comprehensive task management templates following template-driven architecture
- Enhanced all task templates with CRITICAL sections and anti-pattern enforcement
- Integrated surgical discipline principles from fix-master methodology across task system
- Maintained template consistency with relative paths and dynamic scanning compatibility
- CRITICAL: NPM template distribution reliability
- Implemented dual-path template resolution for published vs. dev environments
- Enhanced build to copy
templates/intolib/templates/for runtime discovery - Verified
npm packincludes bothtemplates/andlib/templates/ - Added validation steps to test install from tarball and runtime health
- Updated
README.md,USER_GUIDE.md, andINSTALLATION.mdto reflect v2.3.4 - Documented template distribution strategy at a high level
- ENHANCED FIX-MASTER: Significantly improved
/fb:fix-mastersurgical fix protocol- Refined workflow structure: Clear 7-phase methodology with better decision points
- Agent validation integration: Smart agent selection for complex fixes with explicit criteria
- Enhanced agent workflow: Template-driven agent instructions with focus areas (TypeScript, performance, architecture)
- Structured final reporting: Comprehensive fix summary template with files changed, root cause, testing, and agent feedback
- Improved manual validation: Stronger emphasis on hands-on testing before e2e tests
- Better decision logic: Clear criteria for when to engage agents vs. simple fixes
- Completely rewrote fix-master-enhanced.md with condensed principles and clear workflow guidelines
- Enhanced agent selection process with specific focus areas and review-only instructions
- Improved template consistency and removed redundant content for better usability
- Added structured agent feedback integration with explicit wait requirements
- NEW AGENT: Added
@agent-john-carmackas a full Claude Code agent- Complete dual-layer coverage: John Carmack now available in both persona (
/fb:persona john-carmack) and agent (@agent-john-carmack) formats - Project-aware analysis: Agent automatically gathers full project context via
flashback agent --context - Performance-critical systems expertise: Game engine principles, hot path optimization, functional programming discipline
- Template consistency: Agent template follows established YAML frontmatter and context-gathering patterns
- Complete dual-layer coverage: John Carmack now available in both persona (
- Enhanced dual-layer AI system completeness with all 20 specialists available in both formats
- Created comprehensive john-carmack agent template with YAML frontmatter and context-gathering instructions
- Agent template includes performance-critical systems analysis focus and game engine methodology
- Full integration with existing agent infrastructure and template distribution system
- Maintains consistency across all 20 specialists in dual-layer architecture
- AUTOMATED INSTALLER: Added
scripts/install.shfor single-command installation and updates- Smart template refresh: Automatically updates existing projects with new personas and agents
- Prerequisites validation: Checks Node.js 18.x-22.x and npm 9.x+ compatibility before installation
- Shell integration: Auto-detects Zsh/Bash and creates flashback command alias
- Update detection: Intelligently handles both fresh installs and updates with different workflows
- Project discovery: Finds and refreshes all existing Flashbacker projects automatically
- Enhanced installation experience with single
curl | bashcommand - Comprehensive error handling and user feedback throughout installation process
- Installation documentation now features automated installer as the recommended option
- Manual installation remains available as Option 2 for users who prefer it
- Better user onboarding with clearer installation paths and next steps
- Reduced installation complexity from 4 commands to 1 for most users
- Created robust installer script with comprehensive prerequisite checking
- Implemented smart project discovery to refresh templates during updates
- Added shell configuration detection and automatic alias creation
- Enhanced installation architecture supports both fresh installs and seamless updates
- Full backward compatibility with existing manual installation workflows
- NEW PERSONA: Added
john-carmackpersona for performance-critical systems analysis- Performance optimization: Hot path clarity and deterministic performance analysis
- Functional programming discipline: Minimal side effects and pure function advocacy
- Control flow simplification: Shallow conditional logic and consistent execution paths
- Language-agnostic principles: Real-time systems optimization applicable across all languages
- Game engine methodology: Applies proven game engine principles to any performance-critical codebase
- Enhanced specialist ecosystem from 19 to 20 total specialists
- All personas maintain dual-layer availability (slash commands + dedicated subagents)
- Performance analysis capabilities through specialized John Carmack methodology
- Architectural review focusing on hot path optimization and functional patterns
- Code quality analysis emphasizing deterministic behavior and minimal side effects
- System design guidance for predictable performance and reduced bug risk
- Created comprehensive john-carmack persona template with performance-focused analysis patterns
- Template distribution system supports performance optimization workflow
- Full dual-layer system maintains consistency across all 20 specialists
- Enhanced context-gathering capabilities for performance-critical analysis
- NEW SLASH COMMAND: Added
/fb:fix-masterfor surgical code fixes- fix-master protocol: Systematic 5-phase fix methodology (isolation → analysis → implementation → validation → testing)
- Surgical precision: Enforces minimal, targeted changes over broad architectural modifications
- Manual validation first: Requires manual testing and user confirmation before automated testing
- Anti-duplication: Forces code reading to find existing solutions before implementing new ones
- Complete solutions only: No placeholder files or scaffolding - implements working solutions
- Enhanced surgical fix capabilities for precise, maintainable bug fixes
- Created comprehensive fix-master slash command template with surgical methodology
- Template distribution system supports surgical fix protocol workflow
- Full integration with existing slash command infrastructure
- Promotes systematic, precise bug fixing over broad changes
- GPT-5 INTEGRATION: Added gpt5-cursor agent for advanced analysis capabilities
- gpt5-cursor agent: GPT-5 integration via cursor-agent CLI for second opinions and complex problem solving
- Deep research capabilities for comprehensive analysis and bug fixing
- Enhanced analytical power for challenging technical problems
- Expanded specialist ecosystem from 18 to 19 total specialists
- All agents maintain dual-layer availability (personas + dedicated subagents)
- Updated all documentation (README.md, USER_GUIDE.md, INSTALLATION.md) to reflect 19 total specialists
- Enhanced analytical capabilities through GPT-5 integration
- Improved problem-solving workflow with dedicated GPT-5 access
- Strengthened agent ecosystem with advanced reasoning capabilities
- Created comprehensive gpt5-cursor agent template with cursor-agent CLI integration
- Template distribution system supports GPT-5 integration workflow
- Full dual-layer system maintains consistency across all 19 specialists
- Enhanced context-gathering capabilities for GPT-5 analysis
- NEW AGENTS: Added debt-hunter and hallucination-hunter as full Claude Code agents
- debt-hunter agent: Technical debt detection with CLI scanning capabilities and code quality analysis
- hallucination-hunter agent: AI code validation and semantic correctness analysis for non-functional implementations
- Expanded specialist ecosystem from 14 to 16 total specialists
- Both new agents follow established Hybrid AI+Computer Operations Pattern (CLI scanning + intelligent analysis)
- Full integration with context-gathering workflow via
flashback agent --context
- Updated documentation to reflect 16 total specialists across README.md and USER_GUIDE.md
- Both agents leverage existing CLI scanning capabilities while providing project-aware intelligent analysis
- Agents complement existing slash commands (
/fb:debt-hunterand/fb:hallucination-hunter)
- Created comprehensive agent templates with YAML frontmatter and context-gathering instructions
- Verified agent deployment through full build → link → init refresh cycle
- Template distribution system successfully expanded to support additional agents
- All agents now available in dual-layer system (personas + dedicated subagents)
- CRITICAL: Fixed catastrophic init system bug that destroyed entire .claude directories
- CRITICAL: Fixed framework coexistence issues - Flashbacker now safely coexists with SoftMachine and other Claude Code frameworks
- Replaced hardcoded agent lists with dynamic template scanning for infinite scalability
- Made
--cleaninstall non-destructive (only removes Flashbacker parts) - Enhanced gitignore protection for all Claude Code frameworks
- Framework detection system warns when other Claude frameworks are present
- Dynamic template scanning supports unlimited agents/personas without code changes
- Selective cleanup functions preserve non-Flashbacker installations
- Enhanced error handling and validation throughout init system
- Clean install now preserves other Claude Code frameworks completely
- Agent cleanup uses template-based detection instead of hardcoded lists
- Repository references updated for AgentSea organization
- Template-driven architecture maintains zero hardcoding principle
- Framework coexistence tested with SoftMachine integration
- Security audit completed with HIGH confidence rating
- All documentation updated for public release
This is the first public release of Flashbacker under the AgentSea organization.
Critical Fixes: This release fixes catastrophic bugs that could destroy other Claude Code framework installations. All users should upgrade immediately.
New Architecture: Dynamic template scanning replaces all hardcoded file lists, enabling infinite scalability for agents and personas.
Framework Safety: Bulletproof coexistence with other Claude Code frameworks like SoftMachine.