v0.18.1

Fixed

• FootprintTheme wrapper div uses display: contents so its
  box is invisible to the host's layout. Previously the wrapper
  participated in flex/grid layout as a block-level div that auto-
  sized to its content — which silently collapsed descendants that
  relied on flex: 1 / height: 100% to fill the theme root. The
  bug manifested as "tabpanel / card / scroller height = 0" inside
  themed trees; host apps had to add workaround CSS like
  .tab-content > .fp-theme-root { flex: 1; display: flex; } to
  cope. With display: contents, CSS custom property inheritance
  still flows (vars cascade via the DOM, not the render tree) while
  the layout box is removed. Consumers no longer need workaround
  CSS. One caveat: display: contents elements are removed from
  the accessibility tree in Safari <15.4 — the wrapper is purely
  presentational so this is acceptable.

v0.18.0

Added

• <NarrativePanel> "Copy for LLM" now exports the full debug
  bundle when optional runtimeSnapshot + spec props are provided.
  The copied Markdown now includes: rendered narrative, Final Shared
  State, Commit Log (per-stage writes keyed by runtimeStageId),
  Recorder Snapshots (metrics, tokens, instructions, emit events),
  Subflow Results, and the flowchart Spec (topology).
  Previously only the rendered text was copied, which told the story
  but not the payloads. Paste into Claude and ask "why did iter N
  fail?" — the model now has everything it needs.
• ExplainableShell forwards runtimeSnapshot + spec through to
  the Narrative tab's RightPanel → NarrativePanel so the enhanced
  copy bundle works in the standard zero-boilerplate setup.
• Safe JSON serialization helper (safeJsonStringify) in NarrativePanel
  handles circular references + caps output at 500KB so clipboard
  pastes stay responsive on very long runs.

Fixed

• GanttTimeline showed 0ms durations for every stage.
  extractStageTimings in adapters/fromRuntimeSnapshot.ts still
  read the legacy MetricRecorder.data.stages[stageName].totalDuration
  shape, but MetricRecorder now emits
  data.steps[runtimeStageId] = {stageName, duration, ...}
  per-execution. Adapter now iterates data.steps, extracts
  {stageName, duration} per entry, and sums durations by
  stageName so looped stages (e.g. CallLLM × N iters) show
  cumulative wall time. Legacy data.stages shape still accepted
  for back-compat with older snapshots.

v0.17.0

Added

• <TraceViewer> — drop-in component that renders an agentfootprint.exportTrace() JSON as a fully interactive Behind-the-Scenes view. Accepts a parsed AgentfootprintTrace object or a raw JSON string; validates schemaVersion === 1; surfaces parse / validation errors via an optional onError callback. Internally a thin shell over toVisualizationSnapshots + <ExplainableShell /> — same composition consumers would write by hand. Drop into any React app to give users a "paste a trace, debug visually" workflow without re-executing the agent.

  import { TraceViewer } from 'footprint-explainable-ui';
  
  <TraceViewer trace={pastedJsonString} fallback={<div>Paste a trace</div>} />

• AgentfootprintTrace + TraceParseError types exported from the main entry. Pin consumers to schemaVersion: 1; future shape changes ship as new schema versions with multi-version dispatch.
• 11 new tests (5 patterns) covering parse + validation surface, JSDOM ResizeObserver polyfill added to test/setup.ts.

v0.16.0

Added

• RightPanel with two modes — Insights and What Happened. Memoized toggle between high-level analytics (recorder views, quality scores) and event-level detail (narrative entries, raw log). Replaces the single-view right column with a mode-switched layout in ExplainableShell.
• Data Trace discoverability — the panel now teaches its own capability. Empty state explains "Backward causal chain — trace any value back to the stage that created it." When frames exist, the header gains an italic subtitle "Every value here was derived from the stages below." — surfacing the differentiator at the moment of first encounter.

v0.15.0

Added

• Three-column layout: Flowchart | Inspector (State + Data Trace) | Insights (Story, Performance, Quality, Cost)
• DataTracePanel — backward causal chain visualization (like Chrome DevTools Call Stack)
• InspectorPanel — State + Data Trace tabs for selected node
• InsightPanel — tabs mode (one at a time) or 2x2 grid mode (all visible when flowchart collapsed)
• CompactTimeline — collapsed dot line, expands to Gantt chart
• Responsive: wide (3-col), medium (2-col, no flowchart), narrow (mobile stacked)
• Insight naming: Narrative→Story, Memory→State, Metrics→Performance

v0.12.1

Added

• 90 unit tests — Comprehensive test coverage for narrative sync (position-based), heading numbering (Stage/Selector/Subflow N.M), entry filtering, and loop edge creation in specToReactFlow.

Fixed

• Stage 4 after Subflow 3.x — Root counter increments for subflow step so subsequent stages number correctly.

v0.12.0

Added

• Narrative headings — Type-specific headings: Stage N, Selector N, Decider N, Subflow N.M. Stage numbering accounts for selectors and subflows (Stage 1 → Selector 2 → Subflow 3.1/3.2 → Stage 4).
• Position-based narrative sync — Replaces set-based stageId matching with sequential position mapping. Correctly handles loops (same stageId at different iterations) and parallel subflows.
• Subflow filtering — Main narrative hides subflow-internal entries (stages, steps inside subflows). Shows Entering markers, hides Exiting markers. Internal stages appear in drill-down view only.

Fixed

• Flowchart renders for all specs — No longer requires subflows. SubflowTree sidebar only shows when spec has subflow nodes.
• Loop back-edge rendering — Fixed edge target to use stageId instead of name. Dotted loop arrow now appears in the flowchart.
• Stage 4 numbering after subflows — Root counter increments for subflow step so subsequent stages number correctly.

v0.11.5

Changed

• Narrative hides subflow internals — StoryNarrative filters out subflow-internal entries (stages, steps, conditions inside subflows) and Entering/Exiting markers. The main narrative shows only root-level entries + [Selected]/[Parallel] markers. Internal stages appear in the drill-down view. Future count accurately reflects visible entries only.

v0.11.4

Added

• Subflow group headers in narrative — StoryNarrative shows subflow ID headers when subflowId changes between entries. Subflow entries are grouped consecutively and indented for readability in parallel execution flows.

v0.11.3

Fixed

• Narrative sync with stageId + subflowId — StoryNarrative now matches entries using stageId, subflowId, and stageName against the revealed snapshot set. Entries without identifiers belong to the previous revealed section. Works correctly with selectors, forks, parallel branches, and subflows.
• NarrativeEntry type includes stageId and subflowId — Matches footprintjs v0.17.2 CombinedNarrativeEntry shape.