中文版 | English (default)
lorainwings-plugins
A Claude Code plugin marketplace — spec-driven autopilot orchestration and parallel AI engineering control-plane.

Plugins
| Plugin | Version | Description |
|---|
| spec-autopilot | 5.3.3 | Spec-driven autopilot orchestration for delivery pipelines — 8-phase workflow with 3-layer gate system and crash recovery |
| parallel-harness | 1.3.0 | Parallel AI engineering control-plane — task-graph scheduling, 9-gate system, RBAC governance, cost-aware model routing |
| daily-report | 1.2.8 | Auto-generate and submit daily work reports from git commits and Lark chat history |
Quick Install
# 1. Add marketplace
claude plugin marketplace add lorainwings/claude-autopilot
# 2. Install spec-autopilot (project-level)
claude plugin install spec-autopilot@lorainwings-plugins --scope project
# 3. Install parallel-harness (project-level)
claude plugin install parallel-harness@lorainwings-plugins --scope project
# 4. Install daily-report (project-level)
claude plugin install daily-report@lorainwings-plugins --scope project
# 5. Restart Claude Code
What is spec-autopilot?
spec-autopilot is a Claude Code plugin that automates the full software delivery lifecycle: from requirements gathering through implementation, testing, reporting, and archival.
Key Features
- 8-Phase Pipeline — Requirements → OpenSpec → FF Generate → Test Design → Implementation → Test Report → Archive
- 3-Layer Gate System — TaskCreate dependencies + Hook checkpoint validation + AI checklist verification
- Crash Recovery — Automatic checkpoint scanning and session resume
- Anti-Rationalization — 16 pattern detection to prevent sub-agents from skipping work
- TDD Cycle — RED-GREEN-REFACTOR with deterministic L2 validation
- Requirements Routing — Auto-classify as Feature/Bugfix/Refactor/Chore with dynamic gate thresholds
- Event Bus — Real-time event streaming via
events.jsonl + WebSocket
- GUI V2 Dashboard — Three-column real-time dashboard with decision_ack feedback loop
- Parallel Execution — Domain-level parallel agents with file ownership enforcement
- Modular Test Suite — 102 test files with 1245+ assertions
Architecture
graph TB
subgraph "Main Thread (Orchestrator)"
P0[Phase 0: Environment Check<br/>+ Crash Recovery]
P1[Phase 1: Requirements<br/>Multi-round Decision Loop]
P7[Phase 7: Summary<br/>+ User-confirmed Archive]
end
subgraph "Sub-Agents (via Task tool)"
P2[Phase 2: Create OpenSpec]
P3[Phase 3: FF Generate]
P4[Phase 4: Test Design]
P5[Phase 5: Implementation<br/>Serial / Parallel / TDD]
P6[Phase 6: Test Report]
end
P0 --> P1
P1 -->|Gate| P2
P2 -->|Gate| P3
P3 -->|Gate| P4
P4 -->|Gate| P5
P5 -->|Gate| P6
P6 --> P7
style P0 fill:#e1f5fe
style P1 fill:#e1f5fe
style P7 fill:#e1f5fe
style P4 fill:#fff3e0
style P5 fill:#fff3e0
What is parallel-harness?
parallel-harness is a Claude Code plugin that provides a task-graph-driven parallel AI engineering platform. It enables multi-agent orchestration with strict governance, cost control, and quality gates.
Key Features
- Task Graph Orchestration — Decompose complex requirements into a structured DAG with dependency tracking
- Multi-Agent Parallel Scheduling — Execute independent tasks concurrently with strict file ownership isolation
- Cost-Aware Model Routing — 3-tier automatic routing with escalation, downgrade, and budget control
- 9-Gate System — test, lint, review, security, perf, coverage, policy, documentation, release readiness
- Policy-as-Code — Declarative policy rules with path boundaries, budget limits, model tier caps
- RBAC Governance — 4 built-in roles (admin/developer/reviewer/viewer), 12 fine-grained permissions
- Audit Trail — Full event-level audit with timeline replay, JSON/CSV export
- PR/CI Integration — GitHub PR creation, review comments, CI failure analysis via
gh CLI
- Session Persistence — Memory/file dual adapters with checkpoint recovery
Architecture