v1.3.7: Cleanup + Token Optimization

Changes ### Cleanup - Remove PostHog telemetry (dead code) - Remove DeepSeek-TUI adapter (non-OpenCode CLI support) - Update plugin identity to extendai-lab - Rename tmux sessions omo-* to eai-* ### Token Optimization - Memory injection: conditional (every 5 turns) + file caching - Estimated 35% token savings in 12-turn loops ### Bug Fixes - Fix hasUserContent to exclude markdown headers - Memory tests updated for new caching behavior ### Files Changed - 23 files, +136/-1752 lines - 204 tests passing

v1.3.6 — Plan Progress Fix + Start-Work Flow Optimization

v1.3.6

Plan Progress Fix (boulder/state.ts)

• \getPlanProgress\ dual-mode parsing (OMO-compatible):
  • Structured (## TODOs\ / ## Final Verification Wave): counts section-only checkboxes
  • Simple (no sections): counts ALL column-zero checkboxes, no prefix filter
  • Fixes \

v1.3.5 — Plan Mode + Auto-Review Overlay Refactor + Test Fixes

v1.3.5

Plan Mode (New)

• /ol-plan-enter\ / /ol-plan-exit\ commands for switching to prometheus (planner) agent
• \�nter_plan_mode\ / \�xit_plan_mode\ tools for LLM-initiated plan mode entry
• prometheus 5-phase workflow: Interview → Research → Plan → Save → Exit
• Balanced Question tool: checks existing chat first, accepts direct input too
• Read-only mode: write/edit/bash denied during planning
• \�nter_plan_mode\ denied for prometheus (PLANNING level)

Auto-Review Refactor

• Review overlay uses same \EffectiveAgentOverlayManager\ pattern as plan mode
• Removed duplicate
  eviewOverlayActiveBySession\ /
  eviewResponsePendingBySession\ state
• \session.prompt()\ no longer sets \�gent: 'reviewer'\ — overlay handles routing

Plan Progress & Start-Work Flow

• \getPlanProgress\ dual-mode parsing (OMO-compatible):
  • Structured (## TODOs\ sections): counts section-only checkboxes
  • Simple (no sections): counts ALL column-zero checkboxes
  • Fixes \

v1.3.3

Release v1.3.3\n\n- Added PTY-aware terminal routing with clear short-sync / PTY-recommended / PTY-required classes\n- Added bash-timeout recovery guidance so agents continue after timeout instead of stalling\n- Improved HTML Hub and workspace routing, including nested/worktree filtering\n- Simplified dashboard and config surface for higher-frequency user workflows\n- Fixed workspace path normalization to avoid Linux case-folding issues

v1.3.2

Release v1.3.2\n\n- Added chem-orchestrator primary agent and chem-analyst alias\n- Tightened Prometheus bio-classification behavior\n- Fixed multi-workspace viewer/server routing\n- Expanded bio skill bundle resource loading and reuse guidance\n- Stabilized subtask/mode/context routing behavior

v1.2.3 — Skills 加载修复

v1.2.3 — Skills 加载修复

Bug 修复

• CRLF 行尾导致 skills 解析失败：修复了 Windows CRLF 行尾文件的 frontmatter 正则匹配问题，影响 html-templates、html-deck、academic-tools 等所有 template skills 的加载
• 工具描述修正：修正 load_bio_skills 中 rna-seq → rna-quantification 的类别名
• 系统提示词注入：添加 skills_tool_guide 明确区分内置 skill 工具和 load_skill_template/load_bio_skills

验证结果

方法	工具	状态
方法1	skill（内置）	✅
方法2	load_skill_template	✅ 86 个技能
方法2	load_bio_skills	✅ 42 个技能

影响范围

修复了 4 个文件的 frontmatter 正则：

• src/template-skills/catalog.ts
• src/template-skills/loader.ts
• src/server/index.ts
• src/interview/document.ts

v1.2.2 — Enhanced Plan Logic & New Skills

v1.2.2 — Enhanced Plan Logic & New Skills

New Skills (from Superpowers)

• test-driven-development: RED-GREEN-REFACTOR cycle
• systematic-debugging: 4-phase root cause analysis
• brainstorming: Socratic design dialogue
• receiving-code-review: Technical rigor for review feedback

Enhanced Skills

• plan-protocol: Task granularity standard (2-5 min), two execution modes
• auto-review: Evidence-based completion verification
• team-mode: Two-stage review pattern

New Tools

• save_change: Change proposals to .opencode/extendai-lab/changes/
• save_explore: Exploration notes to .opencode/extendai-lab/explore/
• save_scratch: Session scratch notes to .opencode/extendai-lab/sessions//

Enhanced start-work

• Use question tool for plan selection (prevents agent switching)

Directory Structure

• plans/ — Long-term plans (existing)
• changes/ — Change management (new)
• explore/ — Exploration notes (new)
• sessions/ — Session scratch notes (new)

v1.2.1 — Academic Pipeline & Multi-Format Citations

v1.2.1 — Academic Pipeline & Multi-Format Citations

Academic Writing Enhancement

• Multi-citation format support: APA 7th, IEEE, Chicago, MLA, Vancouver, GB/T 7714-2015
• Multi-paper structure support: IMRaD, Literature Review, Theoretical Analysis, Case Study, Policy Brief, Conference Paper
• Mode selection: Academic Mode (full rigor) vs Engineering Mode (streamlined)

New: Academic Pipeline (6-stage workflow)

• Stage 1: RESEARCH — Literature gathering and source verification
• Stage 2: WRITE — Structured draft with proper citations
• Stage 3: REVIEW — Quality check and improvement suggestions
• Stage 4: REVISE — Address review feedback
• Stage 5: FINALIZE — Generate output in target format
• Stage 6: SUMMARY — Document process and archive materials

7 Pipeline Agents

• research_coordinator — Manages research process
• source_verifier — Checks DOI, journal legitimacy
• outline_architect — Creates paper structure
• quality_reviewer — Assesses writing quality (0-100 score)
• citation_manager — Handles citation formatting
• format_selector — Chooses output format
• process_documenter — Records writing process

Integrity Verification

• Pre-writing checklist (research question, sources, no fabrication)
• Post-writing checklist (valid citations, no plagiarism, claims supported)
• Red flags detection (suspicious results, non-existent citations)

Skills Documentation

• Added comprehensive skill registration methods documentation (docs/skills.md)
• Method 1 (configSkills.paths): 20 skills directly exposed
• Method 2 (load_skill_template/load_bio_skills): 694+ skills via tool loading
• Updated load_skill_template tool description

Scientific Skills Integration (from v1.2.0)

• 617 bioinformatics skills (was 442), 88 categories (was 64)
• Integrated science-skills (37) and scientific-agent-skills (141)
• 24 new categories including quantum-computing, drug-discovery, lab-automation, etc.

v1.1.8: Checkpoint Template Fixes

v1.1.8 — Checkpoint Template Fixes

Bug Fixes

• Checkpoint template must use Write tool
• Added session ID resolution
• Resume template fixes

Verified Working

• Auto-compaction checkpoint works
• Manual checkpoint works
• Both coexist correctly

v1.1.4 — MCP Port Fix

v1.1.4 — 2026-05-27

Fixed

• MCP server port conflict: Changed from fixed port 25569 to dynamic port (OS-assigned). Multiple OpenCode windows can now run simultaneously without port conflicts.