菜单
Discussion mode. Everything said during this mode is treated as thinking material only — no documents are touched. Use for brainstorming directions, or to talk through concerns about a proposal before deciding how to change it.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Interrogates preference-driven decision points. Works through each [TBD] in research.md one by one as numbered plain-text questions; answered items are moved to
Spec-driven development workflow overview. Load this skill when the user says "spec this first / draft a proposal / design first / write a proposal", or when a task is >150 lines / spans 3+ files / introduces a new dependency / involves an architecture choice — to learn the sdd plugin's 11 commands, artifact map, and Shared Principles (HARD GATE / interrogation / Stuck Protection / Anti-Cheating).
Implement the code, advancing by proposal/tasks. A pre-command hook checks that proposal.md carries the required prerequisites. Incremental verification: call $spec-verify close to each node as it lands, don't save it all for the end.
Dispatches a researcher agent to survey industry practices and key constraints, writing findings into research.md (single file; Open[TBD]/Decided maintained as before). Changing direction = snapshot the current research.md into the research/ discarded-draft pile first, then overwrite with the new direction; old drafts can be resurrected at any time.
Verifies the current change by dispatching the independent spec-verifier agent (fresh context — the implementing conversation never audits itself) across three dimensions + charter audit. Every run updates the verification ledger spec/changes/<name>/verify.md (stable finding IDs + round diffing + unfixed-escalation). Can be re-run independently.
Archives the current change to spec/archive/YYYY-MM-DD-<name>/. Only invoked when the user explicitly says "archive". Writes an archive-stage retrospect, checks for uncommitted code, and is guarded by the check-archive hook.
| name | spec-chat |
| description | Discussion mode. Everything said during this mode is treated as thinking material only — no documents are touched. Use for brainstorming directions, or to talk through concerns about a proposal before deciding how to change it. |
Topic: $ARGUMENTS
Everything the user says during this mode is treated as thinking material — no documents are touched.
May Read existing artifacts for discussion context.
Makes "discussion mode" an explicit state. Prevents Claude from automatically editing documents when you share feedback — this is the SDD workflow's "safe pause zone".
The user issues a clear command to switch to another mode:
| User says | Transition to |
|---|---|
| "update the proposal based on what we just discussed" / "revise section X" | $spec-revise [section] |
| "research X again" / "look into Y" | $spec-research <direction> |
| "we just surfaced N TBDs — ask me about them" | $spec-ask |
| "let's draw the architecture" | $spec-design |
| "let's pick this up tomorrow" / "leave it for now" | stay in chat mode |
No explicit switch command → stay in chat mode and keep talking.
$spec-ask's job)| Command | Touches proposal | Touches research | Signal to Claude |
|---|---|---|---|
$spec-chat | ❌ | ❌ | "just talk, don't act yet" |
$spec-ask | ❌ | ✅ (TBD→Decided) | "structurally ask me about preference-driven decisions" |
$spec-revise [seg] | ✅ (partial) | ❌ | "I know what I want to change in section X" |
$spec-propose | ✅ (full) | ❌ | "write or rewrite from scratch" |