openspec-long-running-harness
OpenSpec Long-Running Agent Harness
概述
整合 OpenSpec 规范与 Anthropic "Effective Harnesses for Long-Running Agents" 最佳实践，为跨多个 context window 的 Agent 任务提供：
- 状态持久化 - 通过
feature_list.json和session_state.json跨会话保持进度 - 增量进展 - 每个 session 只处理一个 feature，明确记录进展
- Gate Enforcement - 会话结束前必须验证：工作区干净、新 commit、E2E 通过
- 并行执行 - 支持 3-6 个独立子任务并发，依赖链串行
触发条件（必须命中）
满足任一条件就必须调用本技能：
- 用户使用
/opsx:apply或/openspec:apply且项目存在openspec/harness/目录 - 用户明确要求 "long-running" 或 "harness" 模式
- 用户提到 "跨会话"、"持久化状态"、"增量进展" 并涉及 OpenSpec 工作流
- 项目已初始化 OpenSpec 结构，需要实现多个 feature
More from rexleimo/rex-skills
openspec-parallel-agents
Use when OpenSpec workflows need dependency-aware parallel subagents across OPSX commands, legacy openspec commands, and Codex CLI prompt aliases.
10anthropic-1p-prompt-optimizer
Optimize, rewrite, and evaluate prompts using the Anthropic 1P interactive prompt-engineering tutorial patterns (clear/direct instructions, role prompting, XML-tag separation, output formatting + prefilling, step-by-step “precognition”, few-shot examples, hallucination reduction, complex prompt templates, prompt chaining, and tool-use XML formats). Use for 提示词优化/Prompt优化/Prompt engineering, rewriting system+user prompts, enforcing structured outputs (XML/JSON), reducing hallucinations, building multi-step prompt templates, adding few-shot examples, or designing prompt-chaining/tool-calling scaffolds.
7hybrid-executor
当需要利用 Claude 的代码重构能力或 Gemini 的超长上下文分析时使用。
6spec-kit-parallel-orchestrator
Parallel orchestration skill for Spec Kit workflows. When users mention spec kit, enter /speckit.* or /prompts:speckit.* commands, request spec kit workflow prompt, or need multi-agent collaboration, this skill must be triggered. It validates stages and dependencies first, then splits tasks into 3-6 parallel subtasks, aggregates results, and recursively proceeds to the next round.
6ccg-workflow
Multi-agent orchestration workflow for software development. Coordinates Claude Code (Coordinator), Codex CLI (Backend Authority), and Gemini CLI (Frontend Authority) through slash commands (/ccg:*). Supports propose, explore, new, continue, ff, apply, verify, archive, status, config commands. Schema-driven artifact dependency graph with parallel invocation, session reuse, cross-validation, and trust-based conflict resolution.
1