Files
PaperClipAI/tests/perf/issue-detail
DottaandClaude Opus 5.5 fc9e4cbbb3 perf(ui): show a returning task's saved chat before any server response (#21671)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work
> - The board UI opens a task as a chat. A returning browser restores a
saved read cache from local storage before the first render.
> - #21650 made a returning hard load share the one chat bootstrap. But
the saved cache held no chat, so the first chat frame still waited for
that response.
> - The owner approved saving recent comments in the browser on
2026-11-14, so that cached tasks show their chat at once.
> - This pull request saves the first chat frame of the 20 most recently
opened tasks, with size, age, revocation and isolation guardrails. The
restored chat shows at once and revalidates through the bootstrap.
> - It also adopts the hard-load bootstrap at app start and sends the
other live-run readers through it, so readers that start before the task
page mounts do not send their own requests.
> - The benefit is that a hard load of a task the user opened before
shows its conversation before any server response.

## Linked Issues or Issue Description

**Describe the enhancement**
A hard load of a task that the user opened before still waits for one
server round trip before its chat appears. The saved read cache did not
include comments.

**Current behavior**
With every API response delayed by 800 ms, a returning load shows the
chat at 892 ms (median). The frame waits for the chat bootstrap
response.

**Proposed behavior**
The saved read cache also keeps the first chat frame of recently opened
tasks. A returning load shows that frame before any response, then
revalidates through the bootstrap.

**Why**
The owner approved saving recent comments in the browser for this on
2026-11-14. Cold and cached loads must both be as fast as possible.

Related: #21650 (returning load shares the chat bootstrap), #20594 and
#20591 (returning inbox from the saved cache), #20592 and #21557 (cold
task load from one bootstrap).

## What Changed

- `ui/src/lib/persisted-query-cache.ts`: saves the reads the first chat
frame waits for (`taskChatFirstFrameRef`): recent comments, their input
facts, recent activity, the recent run window, interactions, attachment
metadata, live and active runs, work products and the plan document.
- Only for the 20 saved recent tasks, in the record of the task's own
user and company.
- At most 50 comments per task (the newest; `hasMore` is set), 192 KiB
per task, 1 MiB in total, newest tasks first.
- If storage refuses a record, the record is saved again without chat,
so chat never displaces the rest of the cache.
- All other field exclusions (tokens, credentials, environment values,
embedded documents) still apply. Only `comments` is allowed, and only
for these keys.
  - Expiry is the same as every other entry: 24 hours after the read.
- A 401, 403 or 404 on the task or on any of its chat reads removes the
task and its saved chat from memory and storage at once.
- `ui/src/lib/issue-chat-bootstrap.ts` and `ui/src/main.tsx`:
`adoptHardLoadChatBootstrap` adopts the bootstrap that the HTML started,
at app start, under the task's restored id and identifier. Readers that
start before the task page mounts share it.
- `IssueDetail.tsx`, `LiveRunWidget.tsx`, `IssueRunLedger.tsx`,
`useTaskPluginPanels.ts`: their `live-runs` reads share the loading
bootstrap. Before, whichever observer mounted first could send its own
request.
- `doc/DEVELOPING.md`: the read-cache rules now state what is saved, the
bounds, why, and the owner's approval on 2026-11-14.
- Tests: `ui/src/lib/saved-task-chat.test.ts` (guardrails), a new case
in `restored-task-chat-bootstrap.test.ts`, and an always-on perf test:
"a returning task load shows its saved chat before any API response,
then revalidates".

## Verification

Local production build, local_trusted instance, CDP delay on every API
request, fresh browser context per run, JavaScript in the HTTP cache.
Time to the revealed history with the latest message and an editable
composer. Saved read cache present against cleared (other storage kept),
5 alternating pairs each.

| API delay | Build | Cleared | Present | Requests before frame
(present) |
| ---: | --- | ---: | ---: | ---: |
| 800 ms | master `36101ebcb9` | 1,911 ms (1,895–1,920) | 892 ms
(887–898), released by the bootstrap | 33 |
| 800 ms | this PR | 1,922 ms (1,912–1,936) | 571 ms (556–588), before
any response | 36 |
| 300 ms | master | 916 ms (907–920) | 570 ms (564–576) | 37 |
| 300 ms | this PR | 928 ms (910–998) | 586 ms (568–622) | 37 |

- At 300 ms the bootstrap answers before the page mounts on both builds,
so they are the same. At 800 ms the saved chat removes the wait for the
bootstrap: −321 ms against master.
- The new perf test holds every response for 2.5 s. It fails on master
(frame at 2,595 ms, first response at 2,524 ms) and passes with this PR.
- Guardrail tests, red/green: I disabled each guardrail in turn in
`persisted-query-cache.ts` and ran `saved-task-chat.test.ts`. Each
mutant made its test fail, and all 19 tests pass with the code. Mutants:
no chat saved, no 50-comment cap, no 20-task cap, no per-task budget, no
total budget, no retry without chat when storage refuses, no 24-hour
expiry, no revocation on a chat read, chat stored in the wrong company
record, and no field exclusions.
- Sign-out and other-user isolation use the existing per-user records
and their clearing (no new code). The tests cover them for saved chat.
- "Never overwrites newer data" is the query cache's hydrate rule; the
test covers it. "The bootstrap replaces saved data" uses the
restored-entry rule from #21650; the test covers it.
- App-start adoption test: it fails when the adoption call is removed,
and passes with it.
- Focused Vitest: 25 files that use the changed modules, including
`IssueDetail.test.tsx` and `TaskChatThread.test.tsx`: 1,156 passed.
`pnpm --filter @paperclipai/ui typecheck` passes.
- Browser journeys against this build: `layout-stability.spec.ts`,
`plan-startup.spec.ts`, `cold-bootstrap.perf.spec.ts`: 20 passed (2
opt-in benchmarks skipped).

Leftover early reads on staging (one run after #21650 started
`issues/:ref?view=core` and `issues/:ref/live-runs` before the frame): I
could not reproduce them locally. That included sidebar Recents and a
bootstrap delayed to 1.2 s with CDP initiator stacks. Both reads would
go to the server only if no bootstrap was being tracked for the task, or
if an observer used a direct read. This PR removes both openings: the
bootstrap is adopted at app start, and every `live-runs` observer shares
it. I will check the request timeline on staging after deploy.

## Risks

- Privacy: recent comments of up to 20 recent tasks now stay in local
storage for up to 24 hours, or until sign-out. The owner approved this
on 2026-11-14. The bounds, the revocation and the field exclusions are
tested. Saved data is per user and company, and sign-out clears it in
every tab.
- Storage: saved chat is limited to 1 MiB in total. A record that
storage refuses is saved again without chat.
- A saved chat can be up to 24 hours old when it shows. It is marked for
revalidation, and the bootstrap answer replaces it within one round
trip. A refused revalidation removes it.

## Model Used

- Claude Opus 5.5 (`claude-opus-5-5`), Anthropic, used through Claude
Code with tool use (shell, file edits, browser automation through
Playwright and Chrome).

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-14 17:38:28 -06:00
..

Issue-detail performance baseline

Run the repeatable baseline from the repository root:

pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts

The rig starts an isolated seeded Paperclip instance, runs five samples for each scenario/profile, and writes median-ready raw data plus a Markdown table to test-results/issue-detail-perf/.

Scenarios:

  • S1 warm in-app navigation: loads the Issues list, clears the waterfall, and clicks the seeded issue.
  • S2 cold open: creates a fresh browser context and deep-links to the seeded issue.

Profiles:

  • Unthrottled.
  • Fast 4G network with 4x CPU slowdown.

Override the sample count (minimum five) or port when needed:

PAPERCLIP_ISSUE_PERF_RUNS=7 PAPERCLIP_ISSUE_PERF_PORT=3210 pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts

S1 clicks as soon as the list row is visible. Set PAPERCLIP_ISSUE_PERF_LIST_DWELL_MS=1500 to wait on the list first, as a person does. Task lists preload the chat of their top rows after the list's own reads settle, so S1 with a dwell measures a preloaded open and S1 without one measures an open that usually has to read the task.

Outputs include baseline.md, baseline.json, and a Chrome trace for the first run of each scenario/profile. Open *.trace.json in Chrome DevTools Performance to inspect the issue-detail:* user-timing marks.

Set PAPERCLIP_ISSUE_PERF_PROFILE=1 to capture actual Chrome CPU samples and conversation-readiness transitions in one diagnostic sample per scenario/profile. Import the resulting *.cpuprofile in DevTools and inspect *.readiness.json for the task, comments, metadata, activity, runs, runtime, plan and transcript gates. Each gate records changes with detail.ready; use its last transition before content paint when a gate becomes pending after mount. These diagnostic runs include profiler overhead. Run without the flag for before/after latency claims. Readiness marks stay in the local browser and contain only fixed gate names and booleans; the harness enables them before navigation.

For bundle comparisons, build the UI and serve the production assets from an isolated local instance with PAPERCLIP_UI_DEV_MIDDLEWARE=false. The default managed server above uses development middleware, which does not reproduce production chunk loading. Keep the server, database, browser profile and sample count the same for both builds, and avoid concurrent builds or test suites while measuring latency.

Copy the first run's output outside test-results/issue-detail-perf, then reuse its persisted task with PAPERCLIP_ISSUE_PERF_SEED_FILE=/path/to/before/baseline.json and PAPERCLIP_ISSUE_PERF_BASE_URL pointing to that disposable instance. The harness validates the task identity before measuring. Network spans in the trace use recorded request start/end timestamps relative to task navigation; they include completed requests observed during the sample, even after content paint.

Cold direct task load (chat bootstrap)

A hard load of a task URL starts GET /api/issues/:ref/chat-bootstrap from the inline preload script, before the app's JavaScript loads. The chat's first-frame reads share that answer (ui/src/lib/issue-chat-bootstrap.ts) and fall back to their own requests when it fails, lacks a field, or stalls.

cold-bootstrap.perf.spec.ts delays every API response (300 ms by default) so serial request stages show up as wall-clock time. It opens the task by URL in a fresh browser context with the app's JavaScript in the HTTP cache and no app storage. The finish event is the revealed history with the latest saved message and an editable composer. Its request-shape test always runs: one bootstrap request, no per-endpoint first-frame read before the frame, and no such read completing before it. Reads that start after the live socket connects are its catch-up refresh and are excluded. The benchmark is opt-in:

PAPERCLIP_ISSUE_PERF_BASE_URL=http://127.0.0.1:3287 \
PAPERCLIP_ISSUE_PERF_COLD_RUNS=7 PAPERCLIP_ISSUE_PERF_LABEL=candidate \
  pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts cold-bootstrap -g benchmark

It writes cold-bootstrap-<label>.{json,md} with the median, spread, the request count before the frame, the requests that finished last before it, and the first run's request timeline. PAPERCLIP_ISSUE_PERF_API_DELAY_MS changes the delay. PAPERCLIP_ISSUE_PERF_COLD_HTTP_CACHE=1 downloads the JavaScript on every run. Serve a production build (PAPERCLIP_UI_DEV_MIDDLEWARE=false).

Returning direct task load (saved read cache)

A returning browser restores the saved read cache (paperclip:query-cache:v1:*) before the first render, including the core fields and the first chat frame of recently opened tasks (see the read-cache rules in doc/DEVELOPING.md). The restored chat shows at once; its reads revalidate through the bootstrap and share that one request instead of reading each endpoint. The always-on tests check both: with every response held for 2.5 s, the frame appears before the first response, and no per-endpoint first-frame read starts before the frame. The opt-in benchmark runs alternating pairs with the saved read cache present and cleared (other app storage kept) and writes returning-task-<label>.{json,md}:

PAPERCLIP_ISSUE_PERF_BASE_URL=http://127.0.0.1:3287 \
PAPERCLIP_ISSUE_PERF_COLD_RUNS=5 PAPERCLIP_ISSUE_PERF_LABEL=candidate \
  pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts cold-bootstrap -g returning

Closing a page saves its read cache, so the harness clears storage from a separate same-origin page after the warm-up page closes.

Real-instance click harness

Local fixtures cannot reproduce staging's network round trips, server contention or object-store latency. staging-click-harness.js measures the same journey on any running instance, including staging, from the signed-in browser:

  1. Open an Issues list, for example /<company>/issues?view=mine.
  2. Paste the file into the DevTools console (or evaluate it through a browser automation tool).
  3. Run await __taskOpenHarness.series(["ABC-1", "ABC-2", "ABC-3"]).

Each sample reports the time from the row click to a visible composer and to the latest message in the viewport, the requests started before that point, the requests that finished last before it (releasedBy), the slowest and repeated requests, layout shifts, and any movement of the latest message after it first appeared. serverMs comes from the paperclip_api Server-Timing entry that every API response carries, so a long duration with a short serverMs points to the network or browser queueing rather than the handler. A fetch that is still running when the sample ends is listed with pending: true. Run samples serially and record whether each task's data was already cached (a revisit) or not.

The summary labels each sample (sample 1, ...) instead of the task reference. It replaces query values and every path segment that contains a digit with a placeholder. Other path segments can stay, so review the output before you share it outside the instance.

Task layout stability

The normal browser CI suite also runs tests/e2e/task-loading-frame.spec.ts. It seeds a smaller API-only task and checks desktop/mobile frame and composer geometry, title typography, the initial reading position, Classic preference loading and typography, mobile shell handoff, and a stalled font request. It does not require a separate database URL. The rich 200-message fixture below remains an additional local stress journey.

loading-layout.spec.ts uses a retained synthetic task with a wrapping title, 200 saved messages, document and PR cards, a monitor, and a pending confirmation. It records every animation frame during direct and list-based entry on desktop and mobile, including delayed web fonts and workspace preferences. It rejects real headers shown before the initial conversation is ready, loading-frame movement, different header typography, composer displacement, and movement of the initial reading position. The fixture requires the disposable PostgreSQL URL described below. Its agent has interval and on-demand wakes disabled, including when the retained fixture is reused. The suite does not start provider work. Videos, screenshots, and frames.json are retained with each result.

compact-startup.spec.ts keeps a synthetic local reproduction with 24 messages, a plan, three documents and three saved PR cards. It holds the enriched task response, records each visible frame's card count and geometry, and rejects cards inserted after the conversation becomes readable. It also checks document, older-comment and unknown anchors without full comments/activity/run reads. Run it with the same loopback-only URL override below. The fixture URL is saved in test-results/compact-startup-fixture.json for manual inspection and reuse. It also checks a delayed Coordinator preference with Classic enabled, index-based Coordinator list navigation, and the responder identity through reassignment, comment acknowledgement and legacy runtime discovery. The last case uses a deterministic local process, records identity continuity on each frame, and saves startup/running screenshots. It does not call a model provider. Each send fixture uses a unique task title so repeated runs cannot reuse a previous task and its cancelled run through ordinary task deduplication.

layout-stability.spec.ts exercises the real task route, including Inbox navigation, with 160 Markdown comments and a resolved interaction. It delays initial comments, tests failures and retry, tracks a reading anchor across media growth, composer resizing, properties toggles, older-page loading and Back navigation, and covers mobile reduced motion and explicit comment links. The eight scenarios also exercise same-task hash navigation without remounting on desktop and mobile, and stalled native/log reads that expose Retry after the 15-second request deadline. plan-startup.spec.ts opens tasks directly and from the task list, with and without a saved plan, while holding separate plan reads. The core projection must reveal the saved conversation and plan before those reads finish. Separate plan reads are recorded as diagnostics. The hook tests cover fallback reads for list-only snapshots and responses without a plan projection. Videos and per-frame layout.json measurements are written under each test's output directory. The two-pixel assertion measures a logical row's viewport offset; total scroll offset legitimately changes when history is prepended.

Ordinary task opens defer the concealed history and composer until their initial data is ready. The skeleton and scroll/composer containers retain their geometry. Compact tasks reuse saved replies from completed runs. Their transcript reads start when the reader opens older history through More messages or More run output. Saved Plans from completed runs stay visible while their output is deferred or pending. Live runs, failures, yielded runs, explicit anchors, and runs without a saved reply still load their output. A transcript already shown during this visit stays mounted when the run completes. Comment-link navigation prepares content eagerly so the target can be positioned before reveal. Revealed content stays mounted during refresh. The draft scenario holds a background task read, checks the same editor remains mounted, reloads the saved draft, sends one reply, and independently checks its persisted comment.

To run against an existing disposable local test-drive instance:

PAPERCLIP_ISSUE_PERF_BASE_URL=http://127.0.0.1:3102 \
  pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts layout-stability.spec.ts

The suite creates its own company and fixtures. The URL override accepts only loopback origins and rejects remote hosts. Use a disposable local instance, not a shared or production instance. Without the override, the harness starts its own isolated instance as before. Live provider walkthroughs additionally require a configured native Paperclip runner and Codex authentication; deterministic browser fixtures do not substitute for watching an actual provider run.

saved-reply-startup.spec.ts adds a PostgreSQL fixture with 200 long saved agent replies, 200 completed native runs, and a Plan linked to the latest run. It holds transcript requests, checks both task presentations on desktop and mobile, and measures reply positions on each frame. Opening older history must start the transcript reads and preserve the editor and draft. This case is skipped unless PAPERCLIP_ISSUE_PERF_DATABASE_URL points to a disposable loopback PostgreSQL database whose name ends in _test. Use the same database as the local server:

PAPERCLIP_ISSUE_PERF_BASE_URL=http://127.0.0.1:3102 \
PAPERCLIP_ISSUE_PERF_DATABASE_URL=postgres://paperclip:paperclip@127.0.0.1:5432/task_loading_test \
  pnpm exec playwright test --config tests/perf/issue-detail/playwright.config.ts saved-reply-startup.spec.ts

For the live walkthrough, first assign a disposable task to a native Paperclip runner configured with the Codex provider. Its scratch project should contain a small sum.mjs fixture. The script sends a paced job, scrolls up, disconnects and reconnects, steers a follow-up, and records through completion:

PAPERCLIP_LAYOUT_LIVE_URL=http://127.0.0.1:3102/LAY/issues/LAY-8 \
  node tests/perf/issue-detail/live-feed.walkthrough.mjs

Inspect test-results/task-layout/live-acceptance/ for the video, screenshots, per-frame positions, layout shifts, browser timing counters, and reading-anchor measurement. The live script rejects non-local URLs and uses real provider capacity. Review the video as well as the assertions: layout-shift scores alone do not capture programmatic scrolling or replacement flashes.

Long-running agent chats

The separate agent-chat performance rig covers recent-only startup and scrolling history with long-lived conversations. It is opt-in and does not run in this task-detail rig.