Roadmap¶
This page outlines Cub's development direction, from features currently in progress to ideas under research.
Versioning
Features use stable IDs (e.g., [CB] for Circuit Breaker). Version numbers are assigned at release time, not pre-planned.
Completed Features¶
Recent releases that are now available:
| Version | Feature | ID | Description |
|---|---|---|---|
| v0.23 | Live Dashboard | [LD] | Real-time monitoring with cub monitor |
| v0.20 | Guardrails System | [GS] | Institutional memory and error patterns |
| v0.19 | Git Workflow Integration | [GWI] | Branch-epic bindings, checkpoints, PR management |
| v0.18 | Onboarding & Organization | [OPO] | Improved cub init and project structure |
| v0.17 | PRD Import | [PRD] | Document conversion and task import |
| v0.16 | Interview Mode | [IM] | Deep-dive task specification with cub interview |
| v0.15 | Plan Review | [PR] | Review generated plans before execution |
| v0.14 | Vision-to-Tasks Pipeline | [VTP] | The cub prep workflow |
Active Development¶
Features currently being worked on:
| Feature | ID | Status |
|---|---|---|
| Language Migration (Go + Python) | [LM] | Python CLI complete, bash delegation working |
The language migration is modernizing Cub's internals:
- Python CLI - Typer-based CLI with Pydantic models (complete)
- Bash delegation - Legacy commands delegated to bash (complete)
- Go performance layer - Planned for task operations and streaming
Next Up¶
Prioritized features ready to start. Order reflects current thinking but can shift based on community feedback.
[CB] Circuit Breaker / Stagnation Detection¶
Complexity: Medium | Status: Ready
Detect when the loop is stuck without making progress:
- Track meaningful changes per iteration (file diffs, task completions)
- Configurable threshold (e.g., 3 loops without progress)
- Actions: pause, alert, escalate, or auto-abort
- Stale task recovery: auto-reopen tasks stuck
in_progressbeyond timeout
[RA] Re-anchoring Mechanism¶
Complexity: Low | Status: Ready
Prevent task drift by re-reading context before each task:
- Reload PROMPT.md, AGENT.md, and current task spec before each iteration
- Include git status summary in task context
- Configurable anchoring sources
[CHA] Codebase Health Audit¶
Complexity: Medium | Status: Ready
Systematic analysis to maintain codebase quality:
- Dead Code Detection - Unused functions, variables, orphan files
- Documentation Freshness - README validation, docstring coverage
- Test Coverage Analysis - Per-file coverage, test-code alignment
- Consistency Analysis - Naming conventions, pattern consistency
[AED] Advanced Error Detection¶
Complexity: Low | Status: Ready
Improved error identification in AI output:
- Two-stage filtering to eliminate false positives
- Multi-line pattern matching for complex errors
- Categorize errors (syntax, runtime, test failure, etc.)
[SM] Sandbox Mode¶
Complexity: Medium | Status: Ready
Execute cub runs in isolated sandbox environments:
- Provider-agnostic abstraction (Docker, Sprites.dev)
- Isolated filesystem, optional network isolation
- Resource limits (CPU, memory, timeout)
- Easy diff viewing and change extraction
Backlog¶
Features sequenced primarily by dependencies:
| Feature | ID | Depends On | Description |
|---|---|---|---|
| Parallel Development | [PDW] | - | Git worktrees for multi-branch work |
| Dual-Condition Exit | [DCE] | [CB] | Require indicators AND exit signal |
| Fresh Context Mode | [FCM] | [RA] | Clear context between tasks |
| Implementation Review | [IR] | [PR] | Automated post-task code review |
| Receipt-Based Gating | [RBG] | [IR] | Proof-of-work validation |
| Multi-Model Review | [MMR] | [IR] | Cross-model validation |
| Runs Analysis | [RAI] | - | Post-run insights and patterns |
| Verification Integrations | [VI] | - | External service validation |
Research Ideas¶
Features identified from external tools analysis. These need investigation before planning.
| Feature | ID | Source | Notes |
|---|---|---|---|
| Multi-Agent Orchestration | [MAO] | Gas Town | Swarm coordination, 20-30 agents |
| Session Checkpointing | [SC] | Gas Town | Git-backed state persistence |
| Team Knowledge Base | [TKB] | Compound Eng | Cross-project knowledge |
| Agent Personas | [AP] | Gas Town | Specialized agents per task type |
| Workflow Recipes | [WR] | Gas Town | TOML-based reusable workflows |
| Standing Orders | [SO] | Gas Town | Auto-run tasks at startup |
| Convergent Review | [CR] | Original | 5-pass review for convergence |
| Heresy Detection | [HD] | Gas Town | Detect incorrect beliefs in code |
Dependency Graph¶
STANDALONE (no dependencies):
[CB] Circuit Breaker
[RA] Re-anchoring
[CHA] Codebase Health Audit
[AED] Advanced Error Detection
[SM] Sandbox Mode
[PDW] Parallel Development
[RAI] Runs Analysis
[VI] Verification Integrations
DEPENDENCY CHAINS:
[DCE] Dual-Condition Exit -----> [CB] Circuit Breaker (soft)
[FCM] Fresh Context Mode ------> [RA] Re-anchoring
[IR] Implementation Review ----> [PR] Plan Review
[RBG] Receipt-Based Gating ----> [IR] Implementation Review
[MMR] Multi-Model Review ------> [IR] Implementation Review
IDEAS (dependencies TBD):
[MAO] Multi-Agent ------------> [PDW] Parallel Development
[SC] Session Checkpointing ---> [CB] Circuit Breaker (soft)
[TKB] Team Knowledge Base ----> [GS] Guardrails System
Feature Synergies¶
Features that work well together:
| Combination | Benefit |
|---|---|
| [SM] + [LD] | Safe execution with full visibility ("confident autonomy") |
| [CB] + [DCE] | Robust loop termination |
| [GS] + [CB] | Failures feed guardrails |
| [GS] + [RA] | Guardrails are part of anchoring context |
| [RAI] + [CHA] | Runtime + static quality signals |
| [VI] + [IR] | AI review + real-world validation |
Proposing Features¶
Have an idea for Cub? Here's how to propose it:
1. Check Existing Ideas¶
Review this roadmap and GitHub Issues to see if someone has already proposed something similar.
2. Start a Discussion¶
Open a GitHub Discussion with:
- Problem statement - What problem does this solve?
- Proposed solution - How would it work?
- Alternatives - What other approaches were considered?
- Impact - Who benefits and how?
3. Create an RFC (for major features)¶
For significant features, create a detailed RFC:
# RFC: Feature Name [ID]
## Summary
One paragraph explanation.
## Motivation
Why are we doing this?
## Detailed Design
How will it work?
## Drawbacks
What are the downsides?
## Alternatives
What else was considered?
## Unresolved Questions
What needs more discussion?
4. Implementation¶
Once approved:
- Create an issue linking to the RFC
- Fork and implement
- Submit PR with tests and docs
- Address review feedback
Contributing to Roadmap Features¶
Want to help implement a roadmap feature?
- Check the issue tracker - See if someone is already working on it
- Comment on the issue - Express interest and ask questions
- Review dependencies - Ensure prerequisite features are complete
- Start small - Consider implementing a minimal version first
Standalone Features
Features marked as "standalone" in the dependency graph are great starting points - they don't require other features to be complete first.
Version Planning¶
Cub uses semantic versioning:
- Major (1.0) - Breaking changes, major milestones
- Minor (0.27) - New features, backwards compatible
- Patch (0.26.3) - Bug fixes, small improvements
Version numbers are assigned at release time based on what ships, not pre-planned.
Next Steps¶
-
Contribute
Start contributing to Cub.
-
Development Setup
Set up your environment.