Dashboard redesign — design-system primitives, expandable in-flight hero with stage timeline, activity feed

Spec lifecycle system v1 — typed drafts, dev queue, dashboard, session naming

Design-system parity for critique surface + canonical Agent Input grouping + run-ID copy affordance

Legacy-shim sunset + historical input-bundle backfill

Post-batch cleanup + Anthropic cache engagement + protocol follow-ups

Consumption-card backend follow-ups + protocol-violation UI surface

Phase 0 critique section grouping + live timeline rendering determinism

Consumption card visual rework — CcxCard M3 polish + spec-preview rendering

Canonical prompt-pieces registry + per-attachment token tracking — protocol, persistence, and consumption surfaces

Sources & provenance — investigation outcome and per-critique-card surface (incl. Phase 4 Issue/Comment patches)

Cost & token attribution correction + run-detail header total-cost copy affordance

Prompt capture for full-view modals — Initial Brief persistence and modal hydration

Critique aggregation invariants and resolved-view integrity

Phase 4 deadlock — draft extractor body retention + escape-valve breadth for terminal-ledger AGREED

Active-agent gradient pulse · uniform critique card heights · run-id chip in header row 2

Substantive-convergence escape valve — canonical-promote when both AGREED with terminal ledger but artifact hashes drift

Unify run-status derivation — single source of truth across the All-Runs list, the run-detail page, and the orchestrator's exit-code emission

Phase 0 round cards + side-by-side critique modal — close the round-by-round visibility gap for the brief negotiation

Migrate hardcoded JSX fontWeight numbers to --md-w-* tokens (close the bypass gap)

Diagram skill v2.0.0 — mode-aware split + Material mode + how-it-works regen

Run-detail surface rework — agent chips into Timeline pane, narrow critique compaction, M3 segmented phase progress, timeline card chip slim-down

Design-system pollution scrub (post-arc cleanup)

CSS finalization + v1 token block removal + IBM Plex removal (close the migration arc)

Remaining JSX (app / errors / compare / auth / search / shared) v1 → v2 token migration

design-language.jsx v1 → v2 token migration

run-detail.jsx v1 → v2 token migration

Design system v2 canonicalization — promote v2 to single source of truth, archive v1

Changelog — spec rendered in a full-view modal + working screenshots (or none)

Onboarding tour visual rewrite + unified Admin/Settings + server-persisted onboarding state

Critique filter header height parity + responsive compaction + timeline card right-alignment

How-It-Works as a full-page route + click-to-enlarge SVG viewer

Persist item lifecycle events to the transcript + backfill historical runs

How-It-Works overlay + Changelog tab — full content & component rewrite

Turn-modal items panel — provider badge, segment labelling, Claims panel removal

Badge governance — unified chip primitive, vocabulary, and surface-level rollout

Deep Research consumption & cost tracking — collapsed/unfolded redesign, canonical piece aggregation

Deep Research artifact naming, how-it-works diagrams, and timeline full-view

Turn / Cross-review modal cleanup + Timeline-Critique pane visual harmonisation

Deep Research UI — timeline, critique cards with sources, appendix, validate-run CLI

Deep Research protocol — canonical methodology, lifecycle, and prompts

Full-view modals — fixed 92vh height + remove the 360px accordion-body cap

Agent strip — stop wrapping the model id and activity label when the box is tight

Critique cards — bucket correctness, expanded-card scroll, badge cleanup, height parity with Timeline

Modal sizing + phase rail anchoring + button contrast + critique card layout

Modal chrome + overlay JSX M3 token migration

Run-list page M3 token migration

Timeline phase/turn card treatment + RunDetailHeader children M3 token migration

M3 chrome + run-detail header JSX wiring (plus restore live agent strip dropped by spec 0099)

Loading + States + A11y + Light + Responsive verification sweep (cross-cutting polish + verification of all earlier specs in both themes and three breakpoints)

Onboarding tour overlay + ProgressSegs admin (8-step tour over the live app via data-tour-anchor attributes; no redraw of underlying surfaces; ProgressSegs 8-segment per-user track)

How It Works overlay + right-side menu + Changelog (full-screen M3 dialog · sticky right menu · How It Works ↔ Changelog toggle · 9 collapsible sub-sections with inline diagrams)

Agent input panel + PhaseRail + RoundScrubber M3 anatomy sweep

Consumption pane full rework — collapsed + unfolded with sub-rows + uniform width across phases with round chip above card + sticky bottom legend

Timeline pane M3 rework — header chrome + vertical phase rail outside column anchored to header centers + tl-turn variants + single dashed top border on unfold + REPAIR row variant with explainer

Critique pane M3 rework — Bar 1 (title · phase tabs · totals · drift chip) + Bar 2 (kind tabs · agent · status filters) + collapsible status-grouped sections + Σ Summary state + phase-header sizing taller than card headers

QuestionThread + unified item-card family (Q · D · I · C cards with one anatomy — who → when → what → quote, six-word verdict vocabulary, tonal bubble + quote inside bubble, dashed footer)

M3 modal primitive (basic 560 dp + rich 1080 dp, shape-xl, elevation-3, agent-tinted left border)

M3 tabs + top app bar + chrome compaction (run-list 3→2 headers, top-bar layout when viewing a run)

M3 cards + AgentStrip + badge inventory + hover elevation-2 rule + AgentStrip badge sizing/symmetry

M3 atoms — buttons (5 variants) + FAB + icon button + chips (4 kinds) + status pills (canonical OK) + switches + segmented buttons

Material 3 token & foundation layer (palette, type, shape, elevation, state, motion, density, fonts, icons, base)

Phase 4 drafter-engagement gate (close the round-1 sycophantic-APPROVED loophole)

Parser robustness for cross-round Q/A/issue linkage + code-fence awareness (data-integrity overhaul)

Convergence escape hatches for stuck-AGREED loops (canonical-FSD synthesis, stuck-AGREED escape valve, hard ledger feedback)

Stop hiding Phase 3 / Phase 4 timeline rows on errored & deadlocked runs

Cross-cutting polish — every remaining gap from the 2026-05-18 tweak-cycle audit

Consumption tab rework — phase headers above rows, single card per agent, no duplicate compact bar

Agent Input panel — system-prompt fallback, structural hierarchy, modal vertical space

Unified LoadingState primitive — one delightful loading visual everywhere

Robust run-list loading state + spinner visual + hide 5455

Run-list loading state + server-side hidden-runs filter

Cache /api/runs/{id} snapshot by (run_id, latest_event_seq)

Hotfix: disable Fly auto_stop_machines to stop the proxy flap

Hosted UI modal-load perf — warm machine, gzip, immutable cache headers, server-side LRU

Fly VM memory bump to fix UI-server OOM-on-boot

Hotfix: run-detail.jsx parse error (white-screen regression)

Stable-worktree helper — isolate CLI runs from active orchestrator/feature-branch work

Consumption tab agent-card restructure — equal-height, data-top-bars-bottom, wider bars

Agent Input tab rework — rename, reorder, structural restructure

Question/Disagreement render unification + markdown rendering fix

Critique pane structural — filter strip, Phase 4 split, summary copy

Timeline structural pass — PhaseRail, phase headers, card size, collapsibility

Run-detail header — agent strip equalization, phase-tab info hierarchy, remove blocking banner

Run-list & chrome polish (status pills, top bar, top tabs, right cluster)

Brand-icon system + Design-page DNA reskin

Chip vocabulary + code-cluster expansion

Onboarding (3-screen first-time flow) + landing demo capsule

Cross-run dashboards — /compare (two-run side-by-side) + /search (cross-run query)

Keyboard contract + shortcuts overlay + search palette

Modal primitive (single + split) + sub-tabs + RoundScrubber + provider-symmetric SourceCard

Timeline + critique restructure (PhaseRail + ChipCluster + 3-axis filter + DriftCluster + Summary panel + CardHeadline migration)

Run detail header + chrome restructure + ActiveRunChip

Run list -- sort + attention promotion + filter Tabs + URL state + /-bound search

QuestionThread + QuestionRef + AP-01 enforcement

Tab system (3 variants) + table header

Primitive vocabulary — Button, StatusBadge, Chip, RunIDChip, ThemeToggle, Card, AgentStrip + today's-component migration

Consumption tab — content-vs-billing split + output bar + cross-turn lineage

Design-system foundation — tokens, base, a11y, MDI icons, emoji removal

Reconcile-costs reads run-cost data from Supabase (re-enable daily cron)

Always-on cost verification against provider invoices + pricing-version snapshot

Run-detail resilience + Phase 4 sibling-key separation for repair turns

Critique panel + Summary tab + Consumption tab rework + design unification

Full-view shell standardisation + model pill layout

Turn-input semantics + per-turn badges + side-by-side framing

Cross-round ledger + standing-items input + conservative convergence

Critique data integrity — parser coverage, badge wiring, modal load paths, count reconciliation

Critique classification + load-time resilience + sentiment paragraph + tighter cards

Critique rework — Phase 4 answer linkage fix, compact cards, summary tab, timeline-pane re-alignment

Cost-pipeline integrity — preserve metrics on resume, price the cache tier that was actually used, fold tool spend into the headline

Web search audit UI + agent-pill alignment fix

Web search audit foundation + protocol parser fixes + resume hardening + --notion repeatable

Consumption rework + header-placement fix + app-version chip

Critique navigation — first-class Q+D, side-by-side rework, sentiment cards, click-to-highlight

Inputs foundation — universal Input view, Phase 0 split, two-row live header

Phase-2 hash-drift escape, P2 summaries, live-push flag, dual-research-run skill

Consumption-tab follow-ups — tier-lookup window, click-to-expand bars, per-phase web-search count

Timeline UX pass — inline unfold, per-input segments, real context windows, parser repairs

Token-consumption tab — per-turn context-window visualisation

Cross-review inline comments — Phase 4 side-by-side modal

Negotiate inline comments — side-by-side modal with anchored critique cards

How-it-works restructure — chat-lifecycle diagram, phase accordions, v3.5 process map

Visualisation foundations — modal pattern, summary cards, preflight tabs, attachment ingest

Run-detail header pass 2 + compact theme toggle

Compact run-detail header, "How it works" page, and release notes

Admin allowlist UI, profile menu, and landing-page redesign

Google OAuth + email allowlist via Supabase Auth

Fly.io deployment of the UI server with Supabase-backed aggregator

Supabase schema + `--push` CLI for hosted-deployment track

Hosted deployment kickoff handoff package

Render the last round in deadlocked / errored phase-2 timelines

Live data fidelity — round counts, disagreement parsing, terminal-status pills, chip math

Integration kickoff handoff package

Clearer card stats — Phase 1 badges and explicit labels

Run-id pill and timeline card stats

UI polish and navigation

UI bundle integration

UI HTTP server with SSE

UI run aggregator

Frontend handoff package

Rate-limit-aware retry + resume from prior session

Prompt caching to unblock prod-tier rate limit and cut multi-round cost

Web search wiring + prod-tier full-convergence E2E

Phases 3 + 4 — drafting, review, and final document emission

Phase 2 — plan negotiation with caps, repair, and drafter tiebreak

Orchestrator scaffold + Phase 0/1 end-to-end

Engineering workflow — specs, branches, PRs, semver