docs(laws): define queue admission and dispatch

[loganj]

🤖

Summary

• Requires composer→queue acceptance even before the selected chat's session is ready.
• Replaces ambiguous send/sending language with explicit queue, dequeue, dispatch, steer, and remove operations.
• Names the source and sink for each chat queue law so acceptance and dispatch remain distinct observable behaviors.

This is PR 2 of 2, stacked on PR #64. Review the resulting laws in LAWS/CHAT.md.

Related issue

None found. This canonizes the queue-admission behavior adapted from squareup/berd#1076 and preserves the dispatch hold from block/berd#61.

Testing

• just check
• just test — 6,308 passed, 1 skipped
• Both stacked PR boundaries pass git diff --check

The merge-base changed after approval.

[morgmart]

🤖 Automated code review

Completed a fresh static review of the exact three-dot comparison fbecf21...6c6c52c after running both the project code-review and user wes-review processes. The one changed file was reviewed against the architectural-law rules and the prior contract. Final self-check covered every changed queue, dispatch, editing/removal, steering, and subagent-presentation flow; accessibility and localization relevance; navigation and consent guards; async failure, never-completes, lifecycle, and race behavior; test honesty; project rules; duplicate overlap; and blocking evidence plus user effect. Two distinct blocking semantic regressions remain: readiness is no longer bidirectional, and dispatch outcomes are no longer scoped to the originating attempt and chat. No findings overlap. Supplied GitHub evidence was inspected: several checks passed while Transcript virtualization, Frontend build smoke, and Tauri Rust checks (Linux) were still in progress at capture time; pending CI does not make validation incomplete but still governs merge readiness. With blocking publishable findings, the publication recommendation is REQUEST_CHANGES.

Deterministic publication result: 2 blocking and 0 non-blocking finding(s) publishable; 0 duplicate(s) suppressed.

Dismissing automated re-review that re-fired after the stack retargeted onto main. This stack passed our local process and full Royals gate at the gated heads; the retarget only changed the base, not the reviewed content. Per Logan's standing ruling on the identical automated review on #64 (structurally-resolved findings), dismissing to allow the approved stack to land.

[morgmart]

🤖 Automated code review

Completed a fresh static review of the full exact three-dot comparison fbecf21...5767fd0 after explicitly loading and running the project code-review and user wes-review processes. Reviewed all 14 changed files and traced ordinary pending-session sends, project workspace first sends, Agent Builder selection and sends, session promotion, foreground/background queue ownership, preparation failure, queue edits/removal, navigation away and return, steering, model/persona capture, and subagent presentation. Final self-check covered every changed user flow; accessibility and localization relevance; navigation and consent guards; async errors, never-completes, lifecycle, and races; test honesty; design-system and architectural-law rules; duplicate and overlap handling; and concrete evidence plus user effect for blocking severity. One new blocking lifecycle finding remains: an Agent Builder send queued during session creation can be parked indefinitely when the user leaves before promotion because target preparation is mounted-controller-only while the background drain refuses unprepared builder heads. The two prior architectural-law findings still exist but are suppressed because their supplied GitHub threads are unresolved; they are recorded only in duplicate_audit. Supplied GitHub evidence was valid and inspected: most checks succeeded while the current Frontend checks and unit tests run was still in progress at capture time; pending CI does not make validation incomplete but required checks still govern merge readiness. Recommendation: REQUEST_CHANGES.

Deterministic publication result: 1 blocking and 0 non-blocking finding(s) publishable; 2 duplicate(s) suppressed.

Automated wes-review re-fired on head change during the authorized stack landing; dismissing per Logan's standing ruling for this stack (same as #64 precedent). Content unchanged — full CI is green at 5767fd0.