Ratify B+C composer direction (verification defensibility + plan→execute)

[arielconti10]

What

Makes the B+C composer direction official, per the July 3 session handoff:

• Session doc (sessions/2026-07-03-eve-chat-composer-autonomy.md): "Open composer questions" becomes a decision record — B (verification defensibility) + C (plan→execute transition) chosen; A resolved as behavioral badge, not ladder UI; Bare/B/C/D contract variants stay retired unless B+C work surfaces a need.
• docs/research.md §4.5 (new, additive): levels are analysis vocabulary, not Send-time UI. At Send the user sees a behavioral posture badge + defensibility gaps + (for complex tasks) a plan→execute gate. Cross-links the July 3 audits. §4.1–4.4 untouched.
• Canonical vocabulary as typed data (src/views/patterns/dispatch-moment-data.ts, new): three posture labels ("Asks first" / "Edits freely" / "On its own") and five defensibility gaps (stopping condition, verification method, scope boundary, spend cap, risky actions) defined once. DispatchMomentDemo and the pattern page now consume this module; the risky-actions gap (fifth category from the thread) is added to the demo's pre-Send checklist.
• /patterns/dispatch-moment: header reflects the ratified decision, new "Canonical vocabulary" section renders postures + gaps from the shared data, "Next" section points at the upcoming /playground/compound-composer lab.

Why

The dispatch-moment sketch already embodied B+C but the decision wasn't durable — vocabulary was hardcoded per-surface and the session doc still framed the choice as open. This PR is the ratification step; the compound-composer playground (Item 4) follows in a separate PR.

Verification

pnpm run verify passes (lint, typecheck, format, audit-ui, sync-registry check, registry deps, build, registry build). No registry components touched.

  

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
agentic-craft	Ready	Preview, Comment	Jul 5, 2026 4:43pm

[arielconti10]

Closing after the composer-dressing lab (July 6) reached the opposite verdict: term chrome in the composer is rejected. The lab compared Bare / receipt / scope chips / fully dressed / state-aware variants against product audits of Cursor, Codex, and Claude Code, and the conclusion is that the composer keeps only machinery chrome — meters (context ring, branch), pickers (mode, model), and pointers (@-mentions) — while everything term-shaped (parse echo, scope chips, defaults diff, receipt, defensibility strip) stays out. User-stated constraints are advisory prose; rendering them as chrome gives wishes the visual authority of enforced limits.

What survives from this PR's direction:

• The posture badge (decision A) — it's a picker echoing a real setting, still valid.
• The five defensibility gaps as analysis vocabulary — validated against real incident history, they just don't belong at Send.
• The plan→execute gate — as an in-flight pattern, not composer dressing.
• The receipt content — moves post-dispatch, as the run's opening transcript entry.

Verdict is recorded on /playground/composer-dressing.