Add ecd skill: Evolutionary Constraint Development — freeze product meaning before coding starts

[Zyc-Bryce]

What this skill does

ecd (Evolutionary Constraint Development) turns a vague user request into a constrained plan-code-achieve delivery
loop. It enforces a strict semantic freeze before any code is touched — planning owns meaning, coding owns
execution, and closure owns acceptance.

This addresses a common failure mode in AI-assisted development: the model invents product semantics on the fly,
producing code that looks right but behaves wrong.

How it works

Four slash commands, each with a clear ownership boundary:

Command	Owner	Responsibility
/ecd → pre	Main model	Interrogate, diverge, freeze approval target
plan	Main model + subagents	Converge into a code-ready handoff bundle
code	Main model	Execute only from the frozen handoff
achieve	Main model	Evidence-based closure: archive or reopen

The skill mandates independent subagent review at critical convergence stages (D/G/H/J) — the main model cannot
"play" reviewer; an Agent-tool-spawned independent Claude instance must do it.

Skill structure

skills/ecd/
├── SKILL.md                         # YAML frontmatter + full skill definition
├── scripts/                         # CLI helpers (pure Python stdlib + optional pyyaml)
│   ├── ecd.py                       #   Unified CLI entry point
│   ├── scaffold_case_json.py        #   Stage A initialization
│   ├── render_obsidian_bundle.py    #   Full A-J bundle renderer
│   ├── validate_ecl_bundle.py       #   Bundle validator
│   ├── render_code_run.py           #   Run evidence recorder
│   ├── render_achieve_note.py       #   Achieve verdict renderer
│   ├── render_openspec_pack.py      #   OpenSpec format exporter
│   └── render_canvas.py             #   Visual dependency canvas
├── templates/                       # Markdown templates (Lite + Full variants)
├── schemas/                         # Normalized case format schema
├── references/                      # Playbooks, quality bars, subagent protocol
├── docs/                            # Theory / stages / subagents / implementation (EN + ZH)
├── agents/                          # Claude Code agent interface definitions
└── .claude-plugin/                  # Plugin metadata

Why include this skill

1. Demonstrates an advanced workflow pattern — the staged divergence→convergence→constraint-solving loop is a
   novel application of Claude Code's capabilities not covered by any existing skill
2. Showcases the Agent tool for independent subagents — D/G/H/J stages require real independent Claude instances
   for critique, red-team/blue-team, review, and compilation
3. Production-quality reference — includes structured bundle schema, validation scripts, 20+ templates, and
   bilingual documentation (EN + ZH)
4. Self-contained — pure Python stdlib scripts, no external service dependencies

Checklist

• SKILL.md has valid YAML frontmatter (name, description)
• Skill is self-contained in its own directory
• All file references are relative (no local absolute paths)
• Supporting scripts documented (dependencies: Python 3.8+, optional pyyaml)
• Skill follows the Agent Skills spec format
• Includes bilingual documentation (English + 简体中文)

Credit

Based on Evolution Constraint Planner by
@Etherstrings — the original ECL v2 methodology, CLI scripts, templates, and
schemas were created for Codex and have been adapted for Claude Code.

[Zyc-Bryce]

技能简介

ecd（演进约束开发）将模糊的用户需求转化为一个受约束的 pre→plan→code→achieve
交付闭环。核心原则：规划冻结含义，编码忠实执行，闭环基于证据关闭。

四个阶段各有清晰所有权边界，关键收敛阶段（D/G/H/J）强制使用 Agent
工具启动独立子代理进行对抗审查，主模型不能"自己演"审查者。

包含内容

• SKILL.md + YAML frontmatter
• 8 个 CLI 辅助脚本（纯 Python 标准库）
• 20+ 阶段模板（Lite/Full 双版）
• 结构化 bundle schema
• 12 篇 playbook/quality bar/subagent protocol（中英双语）
• 小白入门完全指南 + 完整使用手册

收录理由

1. 展示了一个现有 skills 中未覆盖的高级工作流模式
2. 示范了 Agent 工具在独立子代理场景的最佳实践
3. 可作为 production-quality skill 参考（含校验脚本、双语文档）
4. 零外部依赖，开箱即用