fix(ce-plan, ce-brainstorm): reliable interactive handoff menus

[tmchow]

Summary

/ce:plan and /ce:brainstorm both show the user a menu at the end of their workflow. That final question was falling back to a numbered chat list instead of using the platform's interactive question tool, even though earlier questions in the same session worked fine. Cause: late-loaded reference files pointed back at the main skill with "(see Interaction Method)" instead of restating the rule inline, menus had grown to 6 options (over AskUserQuestion's native-choice budget), and several labels didn't tell the user what picking them would actually do.

After this change, handoff menus fit inside a 4-option budget at every branch, the question-tool directive is inlined at each handoff point, and every label describes the action it takes.

Fixes #571

What changed

/ce:plan handoff — always 4 options:

• Start /ce:work (recommended)
• Create Issue — restored with tracker auto-detected from AGENTS.md and a one-time prompt to persist project_tracker: on first use
• View & share in Proof — renamed from Share to Proof; description now covers read, comment, collaborate, and share
• Done for now

Dropped: Open plan in editor (covered by Proof), Start /ce:work in another session (not reliably achievable from a skill), and Run additional document review as a menu fixture. The last is surfaced contextually now — when the prior document-review pass left residual P0/P1 findings, the handoff mentions them in prose adjacent to the menu and offers another pass.

/ce:brainstorm handoff — up to 4 options:

• Proceed to planning (Recommended)
• Proceed directly to work — only when the direct-to-work gate is satisfied
• Continue the brainstorm — renamed from Ask more questions; description is self-contained ("Answer more clarifying questions to tighten scope, edge cases, and preferences")
• View & share in Proof — gated so the total stays at 4 (doc exists AND direct-to-work gate closed)
• Done for now

Same contextual surfacing for additional document review. Labels avoid first-person voice so "who is the 'I' here?" never comes up.

Shared:

• Inline the question-tool directive at each handoff (AskUserQuestion in Claude Code, request_user_input in Codex, ask_user in Gemini) instead of indirecting to the main skill.
• Rename Share to Proof → View & share in Proof wherever it appears (including universal brainstorming) so the view/collaborate value shows up in the label, not just the description.

Why ≤4 options

AskUserQuestion renders up to 4 options as native single-select choices. Exceed that, and the tool declines silently and the skill emits a chat list instead — which is exactly what users reported for the final handoff question. Trimming to ≤4 at every branch restores consistent rendering alongside the other decision points in the same session.

Test plan

• pipeline-review-contract asserts the new contextual-surface contract for additional document review and the new (recommended) label position — updated in this PR.
• Run /ce:plan or /ce:brainstorm end-to-end in Claude Code and confirm the final handoff renders as a native selector, not chat text.

---

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 57543d03a5

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".