## Thinking Path
> - Paperclip is the open source app people use to manage AI agents for
work.
> - Task pages must make the current instruction and recent conversation
readable quickly.
> - Compact startup avoids reading all earlier messages and run
activity.
> - Every URL fragment previously selected the full-history path, even
for documents.
> - Saved document cards also arrived after the conversation was
visible. PR statistics could add another row later.
> - This pull request keeps anchors compact and reveals the saved
conversation structure together.
> - The benefit is stable reading position without loading unrelated
history.
## Linked Issues or Issue Description
Related: #17593 introduced compact startup. #14667 covers earlier reveal
and scroll work. This fix retains the bounded history contract.
**What happened?**
Opening a task document link loaded all task history. Ordinary cold
opens could reveal messages before saved document cards and plans
arrived. Later PR metadata added a statistics row and changed card
height.
**Expected behavior**
Every URL fragment keeps compact startup. Document links open their
pane. Older-comment links fetch only the target comment. Saved cards
keep their geometry after first reveal.
**Steps to reproduce**
1. Create a task with many comments, a saved plan, documents and saved
PR work products.
2. Delay the enriched task response and open the task. Observe cards
inserted after messages appear.
3. Open a document fragment. Observe full comment, activity and run
reads.
4. Delay PR metadata refresh. Observe each card grow when diff counts
arrive.
## What Changed
- Remove fragment and tab checks from automatic history expansion.
Resolve validated comment UUIDs through the authorized single-comment
endpoint, encode their path components, retain fetched targets across
document navigation, and preserve cancellation/deletion cache updates.
- Include the saved plan, document metadata and local work-product
projection in the core task read. Skip recovery, workspace and
related-task enrichment. Do not refresh providers. Summary queries no
longer select other document bodies.
- Seed saved work-product and plan queries from core, including already
pending or errored reads, without replacing known lists or an
authoritative empty result. Coordinate the bounded recent reads, saved
cards, latest transcript and settled initial pane width before first
reveal. Keep the thread and composer mounted during refresh and explicit
expansion.
- Open continuation summaries in the document pane. Reserve PR
statistics geometry while provider metadata loads.
- Add a reusable synthetic local task, nine browser checks, frame
measurements and video. Extend the PostgreSQL core-route fixture with a
saved PR row and assert its identity, URL, state and stored metadata
survive the response. Update design guidance and capability maps.
## Verification
- Red: visible document/plan cards changed from 0 to 1 to 4; thread
height grew from 690 to 1,432 pixels. Green: all seven saved cards
appear before reveal, with constant thread height and zero per-card
movement in the final reviewed browser run. The final 1,492-pixel layout
includes the reserved PR statistics lines.
- Red: delayed PR diff counts changed all three cards from 62 to 82
pixels. Green: their heights stay at 82 pixels before and after metadata
refresh.
- Nine browser journeys pass: delayed enrichment,
document/comment/unknown/continuation anchors, document clicks and
explicit history expansion with a retained draft, missing-comment retry,
delayed PR statistics, and retained older-comment position after
document navigation.
- 583 focused UI tests passed on `c91fa107ba`. After the plan-cache
follow-up, all 212 affected page/cache/plan-hook tests and UI typecheck
pass, including 14 plan-hook cases. The original 16 focused PostgreSQL
tests passed for the server changes.
- Merge-review red: holding the duplicate stored work-products read
blocked the first reveal past 10 seconds despite core already containing
the saved cards. Both empty and populated core projections now reveal
while that request remains pending. Malformed-comment tests prove no
target request is made; API tests prove path encoding and abort-signal
forwarding.
- The populated core-route test now inserts a real saved PR work product
alongside the plan and documents, then checks the complete saved
projection through HTTP. All ten task cold-open PostgreSQL tests pass.
- Plan-cache review red: an existing pending or errored plan query
prevented the saved core plan from reaching the consumer. Four real-hook
cases failed before the data-presence fix and pass afterward, for both
present and absent plans. A separately fetched newer plan or known
absent plan remains authoritative.
- Further layout red: the initial properties pane reduced visible thread
width from 775 to 742 pixels after reveal. The reveal now waits for two
stable width frames; the browser check includes width, height, card
counts and per-card geometry.
- `pnpm -r typecheck` and `pnpm build` passed for the original change;
UI typecheck and build pass after review fixes. Token, module-boundary
and map checks pass. `pnpm test:run` was stopped after about 15 minutes
in the serialized server lane. No full local test-suite pass is claimed.
On the earlier head, hosted build/typecheck and all Vitest lanes passed;
one E2E shard failed because the agent-configuration clipboard helper
copied page text into its YAML parser. On `c91fa107ba`, Greptile was 5/5
and every E2E shard passed, including the earlier clipboard case. One
unchanged Discord chat-binding test in server shard 8 returned false
after 185 ms; its 16-test integration file passes locally without
changes. The CI cause remains unestablished. On `be436ce492`, both
server shard 8 and E2E shard 7 passed without changing those tests or
their product code. On final head `e357b3bced`, all CI jobs pass,
including full general and serialized tests, all eight E2E shards,
typecheck, build and the aggregate verification gate. Exact-head
Greptile is 5/5 with no findings. The overall status rollup contains 55
successes, two intentionally skipped Storybook checks and a neutral
Paperclip Review. That advisory review ended without a complete
assessment; all previously posted findings are resolved and it added no
current findings.
- Test startup exceeded an initial 5-second deadline once while
development assets and other checks were active. The final browser
checks allow 10 seconds. These runs establish geometry and bounded
requests; they do not establish a reliable latency improvement. An
initial 503 error fixture exercised global reconnect/backoff. The
target-comment failure test uses a terminal 404.
- Browser data is synthetic. Provider metadata is simulated for the
delayed-response check. No staging deployment or live provider execution
is claimed.
## Risks
- Core task reads now include local saved-card queries and the saved
plan body. Other document bodies and all historical run activity remain
outside that path.
- First reveal waits for the bounded current structure. The existing
15-second incomplete-history fallback and inline retry remain available.
- Continuation-summary links open in the document pane in the current
task interface. Classic tasks keep their existing surface.
- No migration or mutation permission change.
## Model Used
OpenAI Codex, GPT-6 family, with reasoning, code editing, shell tools
and browser testing. The exact serving model ID and context-window size
are not exposed in this session.
## 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
---------
Co-authored-by: Paperclip <noreply@paperclip.ing>
