Repository navigation
Plan integration 7: Add the Web architect journey #14
Description
Parent
- Integrate Study Plans across Now, Web architect, and MCP #7 — Integrate Study Plans across Now, Web architect, and MCP
What to build
Add a Plan with architect journey to the Plans view using the existing live console, while retaining the complete manual form as a fallback.
Acceptance criteria
- The Plans view exposes a clear Plan with architect action.
- Successful launch navigates to the existing live console and labels planning purpose correctly.
- The architect receives interview questions, evidence seeds, existing-plan summaries, and optional brain dump.
- Refresh and reconnect preserve planning-purpose context.
- The manual create/edit/checkpoint interface remains fully functional.
- Exactly one addressed console and WebSocket handle the planning session.
Blocked by
- Plan integration 6: Launch planning-purpose agent sessions #13 — Plan integration 6: Launch planning-purpose agent sessions
Definition of Done
- Browser tests cover launch, fallback, conflict, reconnect, cancellation, and structured errors.
- A representative journey proves that no plan exists before the interview creates one.
- Accessibility, loading, empty, and failure states follow existing Web conventions.
- No duplicate terminal, ACP listener, or WebSocket is mounted.
- Web UI documentation and normative requirements are updated.
Activity
docs(council): brief for code review 3 — Phase 3 (#10 ∥ #11 ∥ #13a) at …
test(web): RED — the 'Plan with architect' journey from the Plans view (
feat(web): 'Plan with architect' beside New plan — one start path, on…
docs(spec): web-ui delta 'Plan with architect journey'; study-plans.m…
test(verify): RED — the plan-integration verification script's regist…
feat(verify): scripts/verify/plan_integration.py — the receipt is the…
docs(archify): plan-integration diagram shows the final structure — n…
docs(closeout): draft the #7–#15 close-out — every criterion mapped t…
test(e2e,uat): the plan-journey worlds remove themselves (#15 DoD "no…
docs(closeout): refresh the #7–#15 close-out DRAFT after review 5, th…
feat(web): the architect door carries the learner's brain dump into t…
NetDevAutomate commentedon Sep 18, 2026
Status: met. Two rows the draft of 2026-09-16 left open — the brain-dump handoff and the abandon-mid-flight test — were closed by follow-on item 2 (owner decision D-B) and by the owner's abandonment-contract decision of 2026-09-18. Source:
docs/architecture/plan-integration/receipts/issue-closeout-draft-2026-09-16.md, re-verified againstmainata5b9f903. Legend as on #8;test_web_plan_architect_journey.pyis e2e-marked (12 browser tests today).T: test_web_plan_architect_journey.py::test_plan_with_architect_action_posts_purpose_planning_and_navigates_to_console(Playwright, real server, fake agent;C: 7ad39941)::test_console_is_labelled_planning_and_label_survives_reconnect; UATarchitect_launchlabel "Planning session — study-plan architect"::test_brief_structure_present_not_wording;::test_brain_dump_reaches_the_architect_persona(browser: the textarea's text is posted asbrain_dump, arrives in the persona as### Learner's brain dump, quoted line by line, never the topic, absent fromGET /api/session/state);T: test_session_start_purpose.py::TestBrainDump(contained section, paragraphs kept, clipped with a marker, over-limit 422, focus start ignores it);tests/js/plan-architect-launch.test.js. The manual form's dump is still never decomposed by StudyLoop (docs/study-plans.md, beside the boundary list)::test_console_is_labelled_planning_and_label_survives_reconnect; the persisted purpose after reload: UATarchitect_launchassertsGET /api/session/state.purposepost-reload::test_manual_new_plan_form_still_works;T: test_web_plans.py(27, protected) for the routes. The Web UI's plan view offers create, activate, milestone checkboxes and preview/record checkpoints; it has no edit-fields control — field revision is the Web API'sPATCH(which since 2026-09-17 includes the mission) and the MCPupdate_study_plan. The checkpoint buttons are covered at the route level, not in a browser::test_one_console_one_websocket;tests/js/plan-architect-launch.test.js(one POST, onestudy-session-start,init()twice still one listener)DoD. Browser tests cover launch, fallback, conflict, reconnect, structured errors, abandonment and the pending-launch contract:
::test_abandoning_a_launch_mid_flight_leaves_no_session_and_no_planends the session through the console's End control as soon as the201lands (no live slot, no plan, no label, one start event, at most one WebSocket), and::test_leaving_before_the_launch_lands_leaves_a_session_like_any_other(owner decision 2026-09-18, option b) holds the start POST while the learner navigates away, releases it, and proves the session exists, reattaches on return and is released by End — leaving never destroys, End abandons, there is no cancel for a pending launch (S: web-ui, design §2 rewritten with the retraction). The planning click's wait for the picker's options is bounded at 8 s (review-6 F3,C: 97efcc4a). Accessibility:aria-busy,aria-describedby,role="status" aria-live="polite", screen-reader label on the subject input, visible label on the brain-dump textarea (not audited with a screen reader). The one-question-at-a-time protocol is persona-text-verified by owner decision D-B: every journey uses the fake agent, which proves delivery of the brief, not model adherence.Landed via #20 (2026-09-17) and #22 (2026-09-18). Closing.
fix(web): the session timer's init() runs once per page load