easyflow

Scans project-level agent directories before subagent dispatch, recommends based on phase keywords, lets user choose, and persists results to disk for anti-corruption

Must use this skill when user triggers /ezfl:audit or requests auditing of implementation outputs. Executes Constitution compliance check (injection point D) + 5 scorer scripts, results written to .harness/metrics/<timestamp>-metrics.json.

Must use this skill when user triggers /ezfl:build or requests implementation / execution of tasks.md. Selects implementer subagent, which executes /opsx:apply in its own session — main agent is forbidden from directly writing implementation code.

Must use this skill when user triggers /ezfl:constitution or /openspec:constitution, or requests creation / amendment / viewing of the project constitution. Manages CREATE / AMEND / SHOW modes for openspec/memory/constitution.md.

Must use this skill when user triggers /ezfl:design or requests entering the design phase. Only responsible for design discussion, Reframe, Premise Challenge, and producing pre_design.md.

Must use this skill when user triggers /ezfl:lock, or requests engineering review / locking plan / review architecture / tech review, or describes a non-trivial change (>3 files, new components, cross-module changes) and is clearly in the pre-coding phase. Reviews plans/proposals from an engineering-manager perspective with adversarial rigor (architecture, data flow, edge cases, test coverage, performance), aiming to find holes rather than go through the motions. **Do NOT** use for: simple bug fixes / single-file changes / pure documentation or configuration changes / requesting 'review of already written code' (that's code review, not plan review).

Must use this skill when user triggers /ezfl:propose or requests generating OpenSpec four-piece set based on pre_design.md. This skill connects three things: (1) prompts user at entry to decide whether to create a worktree (non-blocking); (2) locates and validates the pre_design.md produced by the design phase (falls back to user's original prompt if missing); (3) has the main agent inline-invoke /opsx:propose with exit compliance verification. It is the sole hand-off point between design → lock, and also serves as the worktree decision entry.

Must use this skill when user triggers /ezfl:reflect or requests viewing metrics / overrides / improvement suggestions. Aggregates .harness/metrics/ and .harness/overrides.log, outputs a retrospective report.

Must use this skill when user triggers /ezfl:ship or requests shipping / delivering / completing a change. Executes final verification, branch management, worktree merge-back, and after asking the user, invokes /opsx:archive to archive the OpenSpec change directory.

在 subagent 派发前扫描项目级 agent 目录、按阶段关键词推荐、让用户选择，结果落盘防腐化

用户触发 /ezfl:audit 或要求审计实施产出时必须使用本 skill。执行 Constitution 合规检查（注入点 D）+ 5 项 scorer 脚本，结果写入 .harness/metrics/<timestamp>-metrics.json。

用户触发 /ezfl:build 或要求实施 / 执行 tasks.md 时必须使用本 skill。选定 implementer subagent，由该 subagent 在自己的会话内执行 /opsx:apply——主代理禁止直接编写实现代码。

用户触发 /ezfl:constitution 或 /openspec:constitution，或要求创建 / 修正 / 查看项目宪法时必须使用本 skill。管理 openspec/memory/constitution.md 的 CREATE / AMEND / SHOW 三种模式。

用户触发 /ezfl:design 或要求进入设计阶段时必须使用本 skill。仅负责设计讨论、Reframe、Premise Challenge，并产出 pre_design.md。

用户触发 /ezfl:lock，或要求工程评审 / 锁定计划 / review architecture / tech review，或描述了一个非平凡变更（>3 文件、新增组件、跨模块改动）且处于编码前阶段时必须使用本 skill。以工程经理视角对计划/提案做对抗式评审（架构、数据流、边界情况、测试覆盖、性能），目标是找漏洞而非走过场。**不要**用于：简单 bug 修复 / 单文件改动 / 纯文档配置改动 / 已经在编码中要求'评审已写好的代码'（那是 code review，不是 plan review）。

用户触发 /ezfl:propose 或要求基于 pre_design.md 生成 OpenSpec 四件套时必须使用本 skill。本 skill 串接三件事：(1) 在入口提示用户决策是否创建 worktree（非阻断）；(2) 定位并校验 design 阶段产出的 pre_design.md（缺失时回退用用户原始 prompt）；(3) 由主代理 inline 调用 /opsx:propose，并做出口合规校验。是 design → lock 之间的唯一承接点，且兼任 worktree 决策入口。

用户触发 /ezfl:reflect 或要求查看度量 / overrides / 改进建议时必须使用本 skill。聚合 .harness/metrics/ 与 .harness/overrides.log，输出回顾报告。

用户触发 /ezfl:ship 或要求 ship / 交付 / 完结一个 change 时必须使用本 skill。执行终验、分支管理、worktree 合回，并询问用户后调用 /opsx:archive 把 OpenSpec change 目录归档。