Session view resizing and summary layout

Date: 2026-07-23
Status: Implemented and verified

Goal

Make the read-only session view easier to inspect by allowing the document panel to occupy up to half of the available application width and by presenting the persisted session outcome in task-oriented order.

Session panel resizing

Add a dedicated vertical separator between SessionDocumentsPanel and the area to its right. It must not reuse or interfere with the Model activity separator.
On desktop, pointer dragging can resize the session panel from its minimum usable width up to exactly 50% of the measured application container.
Preserve at least 512 px for the area to the right when the container is too narrow for a 50/50 split.
Support keyboard resizing with arrow keys, Home, and End, expose the current bounds through separator ARIA attributes, and persist the chosen width in local storage under a session-panel-specific key.
On layouts where a desktop split is not usable, do not show an interactive divider.

Summary document contract

The harness remains the source of truth for ordering and semantic projection. Both filesystem and repository-backed snapshots must produce the same ordered document bundle:

Original question.
Final SQL.
Data preview.
Revised question.
Assumptions.
Memories.
Remaining useful technical documents, such as schema linking and validation details not already represented by the preview.

Missing artifacts are omitted without changing the relative order of those present.

Final SQL

Render the persisted sql_final artifact immediately after the original question.
Show an always-available copy button with an accessible label.
Copy the exact SQL text to the clipboard and provide visible success/failure feedback without modifying the SQL.

Preview

Derive the preview from the persisted validation report produced during finalization; do not rerun the SQL when opening a session.
Present the preview as its own Markdown/table document immediately after Final SQL.
Keep execution metadata that belongs to the preview, while excluding unrelated validation-report sections from this position.

Revised question and assumptions

Split the persisted question artifact at its Assumptions heading.
Render the rewritten question as real Markdown, with heading and list structure preserved, after the preview.
Render assumptions as their own Markdown section immediately after the rewritten question. If there are no assumptions, omit the section.

Memories and decision filtering

Build one memory list from effective ledger decisions.
List approved entries first (memory_promoted), followed by declined entries (memory_promotion_declined and legacy memory_rejected where present).
Each entry exposes a human-readable subject and detail/rationale; raw decision-type chips are not shown in the memory list.
Remove these decisions from the generic Decisions document because they are shown in Memories.
Never render phase_approved, table_approved, table_promoted, or column_promoted in the session summary.
Other meaningful decision types may remain in the generic Decisions document.

Component boundaries

harness/tht/session/store.py: pure projection helpers for question sections, preview extraction, decision filtering/grouping, and ordered document construction.
frontend/src/shell/SessionDocumentsPanel.tsx: rendering only; receives the ordered documents and adds the session-panel resize width.
A session-panel-specific resize hook owns measurement, bounds, pointer/keyboard behavior, and persistence. Shared pure resize math may be reused from the activity panel where doing so does not couple their state.
SqlViewer owns the SQL copy control so every Final SQL rendering has the same accessible behavior.

Failure handling

Malformed Markdown or legacy decision lines degrade within the existing per-document error boundary and do not blank the session panel.
Clipboard rejection leaves the SQL visible and reports that copying failed.
A validation report without a preview heading produces no Preview document.
Legacy sessions with a combined revised-question/assumptions artifact are split at read time; their persisted content is never rewritten.

Testing

Harness tests pin identical ordering for filesystem and snapshot document builders, preview/question splitting, decision suppression, and approved-before-declined memory ordering.
Frontend tests pin session-panel resizing to the 50% bound, pointer and keyboard behavior, exact section order, formatted revised-question Markdown, the unified memory list, hidden decision types, and clipboard success/failure behavior.
Run the full harness suite, full frontend suite, TypeScript check, production build, Ruff, Pi gate tests, and git diff --check before completion.

4.9 KiB Raw Blame History