P0-2 artifacts-board.js: guard against LCS OOM on large blob diffs; fall back to note-based render when either side exceeds bound.
P1-2 chat-side-drawer.js: flush pending turn on barge-in (user sends before assistant frame flushes), so drawer stays in sync with graph.
P1-3 chat-refresh-throttle.js (new) + renderer.js: coalesce chat + graph refreshes into per-frame flush to prevent visual jitter under high-frequency intervention/refresh events.
P1-4 artifacts.js: history entries bucket by sessionId; cross-session artifacts no longer bleed into the currently-open session's history rail.
P1-5 rubric-fusion-model.js: seedOnce idempotency guard — repeated seed calls no longer duplicate rubric fixture rows (was surfacing as double-counted evolution nodes).
test/code-bugs-batch.test.js (new, 9 assertions) covers all 5 fixes; test/artifact-evolution-board.test.js updated to align with bySession bucket shape.
9 files, +520/-43. Behavioral fixes only, no API surface change.
T-01 chat-session-graph.js: remove redundant classList.add (setAttribute
below already sets 'active'; leftover from lane-chat-triple merge).
T-02 style.css: rewrite two section headers to describe *what* the section
is, not which internal lane produced it (Rubric-driven views / Artifact
panel tabbed views + version evolution).
T-04-partial chat-session-graph.js / chat-side-drawer.js / growth-v2.js:
drop 'feat/chat-triple-view', 'lane-chat-triple', 'task #140' codenames
from file header docblocks — user-facing source hygiene.
4 files, +5/-7, no behavior change.
Four Context-page enhancements (lane-ctx-deep, F1-F4):
- F1 context window breakdown: replace percentage-only card header with
a stacked-bar breakdown of input / cached / output token buckets,
plus a right-side gauge showing the live window occupancy ratio.
- F2 compact Config tab: fold the sprawling profile Config editor into
a Config tab on the Context page card, with the same yml-leaf
ordering as the top-of-window profile picker.
- F3 intervention marker: on the intervention timeline, emit a marker
glyph at each user-intervention row (turn-flow-glyph-style) so the
card scans as a single stream instead of a header + separate list.
- F4 subagent drilldown: when a turn's tool trace hits a subagent, the
Trace panel's Config + Output tabs get a second row of Subagent
Config / Subagent Output tabs immediately below, driven by the same
fold-in-place shape the parent panel already uses.
4 new renderer modules (compact-config-model / context-window-breakdown /
intervention-timeline / subagent-drilldown), 6 new test files (41 tests,
all node --test style), 4 QA shoot scripts for CDP-driven regression
screenshots.
Replace the process-wide adapter-failure WeakSet with a per-call scope bound to the exact AsyncIterable returned by LlmService.stream(). Give every call a unique wrapper so waterfall middleware can reuse an iterable without sharing provenance.
Move agent-loop recovery classification to the model-stream boundary. Only the final adapter behind that exact call can become an agent/request-error; nested llm/stream calls remain ordinary outer middleware failures while preserving the original Error.
Cover nested calls, reused middleware iterables, and end-to-end agent-loop recovery. Update the package and RFC contracts, bilingual pairing record, and generated API and catalog references.
Adds a version-evolution capability to the Artifacts pane:
- Evolution chain: LCS-based diff across successive versions of the
same logical artifact; when a blob is missing, the chain honestly
surfaces a placeholder rather than fabricating diff content.
- Board / Timeline tab switcher on the Artifacts page.
- Auto-grouping into 5 kinds so the Board reads at a glance.
Files:
src/renderer/artifacts-board.js — new Board / Timeline / Evolution views
src/renderer/artifacts.js — wire Board tab + evolution rail
src/renderer/index.html — script tag + tab markup
src/renderer/style.css — tail append (~200 lines)
test/artifact-evolution-board.test.js — 23 new tests
docs/artifact-board-fixture.json — fixture seed
docs/qa-artifact-evolution/ — 3 QA screenshots + fixture.html
scripts/qa-cdp-shoot-artifact-v2.mjs — QA capture harness
Merged as source-repo test-real @ 7279870 (with lane-side conflict
resolution against test-real @ c1d93aa: style.css tail-append union;
independent of Lanes C/A/D CSS sections).
Test suite: 1727/1727 pass (+23 over Lane C+A+D baseline of 1704).
Follow-up L-3 (in upstream ledger): runtime ArtifactServer to grow
a snapshot store so the "missing blob" fallback becomes rare.
Architectural refactor unifying three previously-independent pages
around a single shared event-log model:
rubric-fusion-model.js — the shared model. One log of scored
events keyed by (rubric, runtime, timestamp), plus a small set
of derivations (recent-scores window, per-rubric aggregates,
per-runtime rubric grid) that all three views subscribe to.
rubric-fusion-seed.js — fixture seed used by tests and the
isolation-machine screenshots below.
The three page views:
Rubrics — now leads with a hint card (recent pass/fail streaks,
stalest rubric) and a stats strip (7-run rolling pass rate,
latest score, delta vs. previous). Rubric list is unchanged.
Growth — rewritten. The old compact-window event feed is gone;
the page is now a pure SVG time-series curve over the event
log (score-y × time-x), with per-rubric coloring and a
hover-time tooltip. -420 lines of the old feed → +630 of the
curve renderer.
Runtime — adds a "Rubric grid" tab next to the existing Status
tab. Grid is one row per rubric × one column per runtime;
cells are pass/fail/unrun, click-through to the raw event.
Wiring is careful about the five hot zones the parallel lanes are
each touching (nav / chat / context / trace / artifact) — Lane D
stays entirely inside rubric/growth/runtime and their shared model,
so this composes with the C/A lanes without stepping on their
finishTurnContainer or chat-triple regions. style.css conflict
against the C+A combined append was tail-append vs tail-append and
resolved by concatenation (brace balance verified).
rubric-fusion-model.js 419 +
rubric-fusion-seed.js 624 +
rubric-fusion-model.test.js 181 +
rubric-fusion-views.test.js 189 +
growth-v2.js +631/-420 (SVG time-series rewrite)
rubrics-page.js +89 -0 (hint + stats strip)
runtimes-page.js +198 -0 (rubric-grid tab)
index.html +7 (three mount points)
style.css +252 (new page sections)
rubric-fusion-fixture.json +94 (test/screenshot seed)
Test suite: 1704/1704 pass (+22 over Lane C+A baseline). Isolation-
machine fixture screenshots (rubrics/growth/runtimes,
docs/rubric-fusion-shots/0{1,2,3}-*.png) reproduce from the merged
HEAD.
Adds a three-layer view to the Chat pane:
- List view — turn cards get colored left-edge by role/state
(user/assistant/tool/error), replacing the flat
monochrome stack; readable at a glance for long
sessions.
- Side drawer — a right-side collapsible panel opened via the
Details button on any turn; shows the raw payload,
annotations, and jump-links to the trace tri-view
without leaving the chat.
- Session Graph— a top-level SVG view of the whole session's turn
graph (user → assistant → tool chain → subagent
branch); zooms out from the linear list to the
session's structure.
Renderer wiring is additive: the three new modules mount into two
new hook points in index.html; turn/start now emits data-turn-id
so the drawer can round-trip. finishTurnContainer keeps its Lane C
signal-chip pass and picks up drawer wiring at the tail — the two
tails are independent and compose cleanly.
chat-session-graph.js 198 +
chat-side-drawer.js 263 +
qa-cdp-shoot-chat-triple.mjs 204 +
chat-triple-view.test.js 222 +
index.html 43 +
renderer.js 111 + (wiring + data-turn-id)
style.css +273 -1 (three view sections)
Test suite: 1682/1682 pass (+14 over Lane C baseline). Isolation-
machine fixture screenshots (list/drawer/graph, docs/qa-chat-triple/
0{1,2,3}-*.png) reproduce from the merged HEAD.
Known non-blocking niceties, tracked as follow-ups (out of scope
for this commit):
- Details button and existing Context Rail both target the
right-side slot; a small drawer-mutex pass unifies them.
- Session Graph fixture includes an extra subagent turn beyond
what the list view shows.
Adds a small signal-detection layer over trace payloads and surfaces
three kinds of semantic signals inline in the trace tri-view (Timeline
/ Graph / turn footer chips) so a researcher can spot repetitive
tool-call loops, redundant identical calls, and plan-completion beats
without hand-scanning the raw log:
- loop-detected — three or more repeats of the same tool.method
signature inside a window, badge in Timeline gutter
+ colored ring on the Graph node
- redundant-call — identical (tool, arg-hash) call twice without any
state change in between, badge + subdued styling
so it stays advisory not alarmist
- plan-* chips — plan/subplan step-complete beats surfaced above the
assistant body in the turn footer, wired through
the finishTurnContainer tail
Pure rendering; no wire-format changes needed on the runtime side —
signal detection is fixture-driven off the same payload the tri-view
already consumes. Companion RFC L-2 in docs/upstream-ledger.md asks
the runtime to emit these signals natively rather than deriving them
in the shell, so once L-2 lands the shell will consume upstream signals
and this detector becomes a fallback.
Files:
src/renderer/trace-signal-detect.js new (loop/redundant/plan)
src/renderer/trace-tri-view.js +24 (wire detector output)
src/renderer/trace-timeline.js +39 (badge in gutter)
src/renderer/trace-graph.js +30 (colored ring)
src/renderer/renderer.js +81 (finishTurnContainer chips)
src/renderer/index.html +1 (one script tag)
src/renderer/style.css +58 (badge/ring/chip rules)
docs/upstream-ledger.md +121 (append L-2 RFC)
scripts/qa-trace-signals-fixture.mjs new (headless SVG proof)
scripts/qa-trace-signals-shoot.mjs new (CDP live shoot)
docs/trace-signals-shoot/*.html new (3 fixture-driven shots)
test/trace-signal-detect.test.js new (detector unit tests)
test/trace-signal-overlay.test.js new (renderer overlay tests)
Test suite: 1668/1668 pass (17 new). Fixture-driven signals-01-timeline-
loop.html regenerated byte-identical from the merged HEAD.
path.resolve('') is the process cwd, so an empty configured cwd would
silently reintroduce the launch-directory fallback the parent-session
cwd resolution removed. Fail at plugin load with an actionable message
instead.
The ACP backend resolved an unset cwd to the harness process's launch
directory, so a delegated child ran — and announced its ACP session —
in the server's cwd instead of the delegating session's workspace.
Resolve the working directory explicitly: the config cwd override
(made absolute and validated at load), else the parent session
header's cwd (validated at start), failing loud before spawning when
neither exists.
Condense the app-bundle sentence after combining the current-master CLI wiring with the recovery architecture changes. The text still identifies the TUI, line-oriented, headless, and ACP front doors and their output contracts.
This restores the enforced architecture.md word ceiling without raising the budget or relocating an architecture-level seam fact.
Document every context-window path returned by mapStopReason: recognized provider error wording, successful stop usage beyond the resolved window, and zero-output length stops that fill the window.
This keeps the public return contract aligned with usage-based overflow classification without duplicating the provider-specific detection table owned by pi-ai.
Use the repository-mandated 回放 terminology for replayable pressure and replace the literal cancellation-owned rendering with idiomatic Chinese that preserves boundedness, cancellation authority, and monotonic retry behavior.
The corresponding English clauses remain accurate. Re-record the English and Chinese pair after checking the corrected text against the implemented recovery contract.
Dispatch RequestErrorDecision and CompactionTrigger through explicit discriminant switches. End each closed union with assertNever so new variants fail compilation instead of silently inheriting fail or pressure behavior.
This preserves the current retry, fail, pressure, and overflow semantics while aligning the new recovery seams with the repository closed-union contract.
Contain arbitrary plugin and runtime failures even when a thrown Proxy traps instanceof checks or its string coercion throws. Fall back to a stable diagnostic instead of letting executeCli reject outside its exit-code contract.
Route abort reasons through the same total renderer so cancellation cannot escape containment through an exotic reason value.
Add a focused regression that exercises both hostile inspection paths and verifies stdout remains empty, stderr remains labelled, and the CLI resolves with exit code 1.
Report context-disposal failure as an independent outcome even when argument, boot, task, or output handling has already produced a primary diagnostic.
Keep the primary error first, append the cleanup error, and retain the nonzero exit status so operators can see both the initiating failure and the possibility that teardown or persistence did not complete.
Add a regression that combines an invalid app composition with a failing disposer and asserts both ordered stderr lines.
Align the one-shot app with the shared agent-spine contract that landed on master after this branch began. Expose maxParallelToolCalls, dshHome, toolBash, and toolTasks through the Loader schema and route them with pickSpineConfig().
This restores deployment control over tool-call concurrency, the shared Harness home, background bash, and task_output wait bounds instead of silently retaining owner defaults.
Exercise all four fields through the composed runtime, document the package-level contract, and regenerate the config catalog from the owning interface.
Yichen Jiang
ZiyaZhang
Tianyi Cui
ZiyaZhang