spike(framework): make the contributor AI environment portable across supported hosts

未关闭
#1,236 0 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
12/100
Issue 类型
文档
描述清晰度
基本清楚
活跃度
活跃
技术栈
markdown, shell, typescript
领域
devtools

调研方向

This is a design spike, not a code change: start from the inventory table and the cli/.claude/rules/00-architecture/0-hexagonal.md and cli/.claude/skills/tools/SKILL.md files it cites, plus scripts/check-hosts.cjs and scripts/dev-sync.sh. Done means a versioned inventory, a host compatibility matrix with reproduction commands, and a recommendation covering scoping, user-owned files, and bootstrap risks, all agreed with maintainers before any PR.

由索引模型根据 Issue 内容生成。

描述

Problem

AIDD's own contributor AI environment is partly Claude Code-specific. Contributors modifying ai-driven-dev/framework with Codex, Kilo Code, OpenCode, or other supported hosts may not receive the same project guidance, architecture rules, skills, and automatic feedback as Claude Code users.

This concerns contributing to the framework itself, not the AIDD plugins distributed to downstream projects.

Static inventory of next at commit 8d67ae942fd61e36daca9b1f28a25ad065ca26c9:

Contributor resource Current state Portability question
Root .claude/rules/ 3 rules No automatic Codex discovery established
cli/.claude/rules/ 12 CLI architecture and development rules No equivalent project-level Codex projection found
cli/.claude/skills/ 5 skills: distribution, framework, test, tools, translate No cli/.agents/skills/ equivalent in the tracked tree
Root .claude/skills/release-video/ 1 skill Symlink under .agents/skills/; live discovery still needs verification
Root .claude/hooks/ PostToolUse and WorktreeCreate hooks No equivalent host behavior demonstrated
cli/CLAUDE.md CLI-specific guidance and memory imports No corresponding cli/AGENTS.md
Root CLAUDE.md and AGENTS.md Overlapping contributor instructions Effective precedence and drift need evaluation

Biome, architecture tests, Lefthook, and CI already enforce many constraints independently of the assistant. However, catching a violation after generation is not equivalent to making the architectural rules available during implementation.

make setup, make reload, and make check-hosts provide multi-host developer tooling, but their current checks do not establish contributor-rule and workflow parity. These are static findings, not proof of a runtime failure for every host.

Scope

  • Inventory all contributor-owned instructions, rules, skills, hooks, scripts, and referenced assets, including nested cli/ resources. Classify ownership, repository scope, and mandatory versus optional behavior.
  • Verify native discovery, instruction precedence, path-scoped rule applicability, skill invocation, supporting-reference resolution, and hook behavior across Claude Code, Codex, Kilo Code, and other supported hosts where feasible. Report tested versions and unsupported cases.
  • Map each rule to existing deterministic checks (Biome, architecture tests, Lefthook, CI), separating agent-time guidance gaps from already enforced invariants.
  • Compare a minimal shared-file/host-projection approach with an optional framework-contributor plugin (provisional name aidd-contributor) using existing AIDD CLI translation, distribution, and installation capabilities.
  • Build a narrow proof of concept using the tools contributor skill and representative CLI architecture rules; test discovery, applicability, reference paths, and repository-scoped behavior.
  • Recommend the smallest justified follow-up, preserving a bootstrap path that works even when the AIDD CLI or plugin installation cannot run.

Acceptance criteria

  • A versioned inventory identifies contributor resources, owners, scopes, and current host entry points, including the 15 rules, 6 skills, and 2 hooks in this baseline.
  • A host compatibility matrix records observed discovery, rule applicability, skill invocation, reference resolution, and hooks, with reproduction commands, host versions, and explicit uncertainty.
  • Documented portability gaps distinguish missing agent-time guidance from deterministic constraints already enforced elsewhere.
  • A reproducible POC compares shared/native projections against the proposed contributor plugin, without assuming either architecture must win.
  • The recommendation addresses repository scoping, user-owned file preservation, updates, duplicate prevention, host-specific hooks, and bootstrap/dependency-cycle risks.
  • Contributors retain a functional baseline through AGENTS.md, CONTRIBUTING.md, make, pnpm, and local tests without relying on the new plugin.
  • The outcome names the smallest necessary follow-up implementation work, if any.

Prior art in this repo

Out of scope

  • Changing end-user plugin behavior or replacing the framework's canonical Claude-format source.
  • Assuming the dedicated contributor plugin is already the chosen implementation.
  • Forcing hosts to expose identical hook mechanisms.
  • Installing repository-specific contributor rules globally in unrelated projects.
  • Removing deterministic architecture gates, automatically migrating user-owned files, or introducing a second independent translation engine.
  • Implementing a full migration or opening a PR before maintainers agree on the proposed direction.
主要语言
TypeScript
星标
481
派生
45
平均合并
1 天 13 小时
30 天内合并 PR
91

环境准备

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

ai-driven-dev/framework 的其他 Issue

查看 ai-driven-dev/framework 的全部 Issue

相似的 Issue

更多 TypeScript Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。