This roadmap outlines the development path from v1.0 to v2.0 and beyond.
Status: ✅ Released
Major Feature: Full upstream sync to Algorithm v1.8.0 with Wisdom Frames, Verify Completion Gate, and 10 new skills
What's New in v2.0:
- 8 Effort Levels — Instant, Fast, Standard, Extended, Advanced, Deep, Comprehensive, Loop
- 25-Capability Full Scan Audit — Replaces Two-Pass capability selection
- Constraint Extraction — Mechanical [EX-N] extraction before ISC
- Self-Interrogation — 5 structured questions before BUILD
- 7 Quality Gates — QG1-QG7 must pass before proceeding
- PRD System — Persistent Requirements Documents for cross-session tracking
- Start Symbol ♻︎ — Replaces 🤖
- Verify Completion Gate (v1.6.0) — Prevents "PASS" claims without TaskUpdate calls
- Phase Separation Enforcement (v1.6.0) — "STOP" markers on every phase
- Zero-Delay Output (v1.6.0) — Instant output before processing
- Effort-Scaled Gates (v1.3.0) — Self-Interrogation, Constraint Extraction, QG6/QG7 scale by effort
- Wisdom Frames (v1.8.0) — Domain knowledge accumulation across sessions (
MEMORY/WISDOM/) - Wisdom Injection (v1.8.0) — OUTPUT 1.75 injects domain wisdom before ISC creation
- BUILD Capability Execution (v1.8.0) — Explicit substep within BUILD phase
- 10 New Skills — Cloudflare, ExtractWisdom, IterativeDepth, Science, Parser, Remotion, Sales, USMetrics, WorldThreatModelHarness, WriteStory
- 6 New/Updated Handlers — algorithm-tracker, agent-execution-guard, skill-guard, check-version, integrity-check, format-reminder (updated)
- WisdomFrameUpdater.ts — CLI tool for managing wisdom frames
- 5 Seed Wisdom Frames — development, deployment, security, architecture, communication
- Security Validator fix — env var prefix stripping (upstream #620)
- Rating Capture fix — 5/10 noise filter
- Symlink support — GenerateSkillIndex follows symlinks
- SessionHarvester — PAI_DIR variable rename
Documentation:
- CHANGELOG.md — Full release notes
- docs/specs/UPSTREAM-SYNC-v1.8.0-SPEC.md — Porting specification
Status: ✅ Released
Major Feature: 16 specialized agents with model tier routing and 3 provider presets
What's New in v1.3:
- 16 Agents - Expanded from ~11 to 16 specialized agents
- Model Tiers -
quick/standard/advancedrouting per agent viaopencode.json - 3 Presets - Anthropic Max, ZEN PAID, ZEN FREE (replaces 8 provider system)
- Rewritten Wizard - Prerequisites → Dev build → 3 presets → Identity → Config
- Researcher Renames -
ClaudeResearcher→DeepResearcher,PerplexityProResearcherremoved - Profile YAML - New format:
default_model:+agents:with optionaltiers:
Status: ✅ Released
Major Feature: Real-time event streaming, SQLite persistence, Vue 3 dashboard
Status: ✅ Released
Major Upgrade: Full PAI 2.5 Algorithm + 5 new handlers
What's New in v1.1:
- PAI 2.5 Algorithm (v0.2.25) - Full 7-phase format with ISC tracking
- Voice Notification Handler - ElevenLabs + Google TTS + macOS say fallback
- Implicit Sentiment Handler - Automatic satisfaction detection from user messages
- Tab State Handler - Kitty terminal tab title/color updates
- Update Counts Handler - Skill/workflow counting at session end
- Response Capture Handler - ISC extraction and learning capture
Status: ✅ Released
What's Included:
- Skills system with 29 skills (CORE, Algorithm, Fabric, Research, etc.)
- Plugin system (security validator, context loader)
- Memory system (projects, sessions, learning)
- Agent system (14 agents, PascalCase naming)
- Skill search and indexing tools
- Full TypeScript tooling with Bun runtime
Goal: Complete remaining upstream items and polish
Features:
- PRD auto-creation via AutoWorkCreation plugin handler
- Dynamic Algorithm version reading via LATEST file
- Enhanced Wisdom Frames with auto-population from session learnings
Goal: Advanced setup options and system health
Features:
- Skill selection UI (enable/disable individual skills)
- System health checks and diagnostics
- Configuration validation
- Migration assistant from Claude Code PAI (interactive)
Goal: Complete feature parity with latest upstream PAI + seamless migration
Major Features:
-
Auto-Migration System
- One-command migration from Claude Code PAI
- Skill mapping and compatibility layer
- Memory import (sessions, projects, learning)
-
Advanced Skill Orchestration
- Skill dependencies and auto-loading
- Parallel skill execution
- Community skill marketplace
-
Enhanced Security
- Sandboxed skill execution
- Granular permission system
- Audit logging
-
MCP Server Adapters
- deepwiki-enhanced (GitHub repo Q&A via Devin API)
- Community MCP server integrations
We value community input! Here's how to shape PAI-OpenCode's future:
- Vote on Features: Comment on roadmap issues
- Propose Ideas: Open a discussion
- Contribute Code: Tackle items from the roadmap (CONTRIBUTING.md)
- Share Use Cases: Tell us how you use PAI-OpenCode
| Version | Release Date | Highlights |
|---|---|---|
| v2.0.0 | February 2026 | PAI v3.0 / Algorithm v1.8.0, Wisdom Frames, Verify Completion Gate, 39 skills |
| v1.3.0 | February 2026 | Multi-Provider Agent System with Model Tiers |
| v1.2.0 | February 2026 | Observability Dashboard + 14 handlers |
| v1.1.0 | February 2026 | PAI 2.5 upgrade + Voice/Sentiment handlers |
| v1.0.1 | February 2026 | Anthropic API fix, ISCValidator improvements |
| v1.0.0 | January 2026 | Initial release - core PAI on OpenCode |
Stay Updated:
- Watch this repo for releases
- Follow Discussions for announcements