DottaandPaperclip
1353a232f2
## Thinking Path > - Paperclip lets people direct agents while their work is active. > - Product E2E checks must prove that saved user input reaches the intended work. > - The existing active chat story depends on a file in a local agent workspace. > - That setup cannot qualify an attached deployment with remote agent execution. > - This change adds one shared active follow-up journey through the normal task UI and API. > - Independent native operation, saved input, final reply, and cleanup checks make the result inspectable across targets. ## Linked Issues or Issue Description **What existing behavior does this improve?** Product E2E coverage for a user follow-up sent while a native agent is working. **Current behavior** The classic chat story holds work open with a local file. The shared deployment-targeted harness has no active follow-up journey. A running heartbeat alone cannot prove that work was active when input arrived. **Proposed behavior** Run one bounded command through the real native provider. Observe its active execution before and after one follow-up is saved. Require one final reply with the late requirement, correct task and agent identity, an empty input queue, and visible readback after reload. Use the same story on managed and attached targets. ## What Changed - Add a shared `followup-while-running` journey with normal task creation and reply submission. - Require a bound native turn and pending shell execution at the input boundary. Retain a closed receipt with source identity and a row hash. - Accept one steered work run or one normal successor (1–2 native runs). Preserve every other journey's declared run contract. - Reuse the saved native final reply observer, run binding checks, usage collection, target lifecycle and private evidence path. - Accept canonical builtin/process execution events independent of provider tool display names; calibrate against real OpenCode, Codex and ACP mappers. Reject Paperclip semantic tool names across native wrappers/prefixes and unnamed or ambiguous builtin calls. - Recognize only the observed legacy PRP v1 builtin/bash/unknown/readOnly=false command shape and fail promptly when owned work finishes before a live boundary can be observed. - Ready the task page/composer before the observed command boundary, use a bounded 90-second command, and preserve active-after-persistence proof. - Add adversarial oracle and retained-result calibration. Register the journey, include its sources in the definition hash, and document its coverage. - Keep six real browser document-readiness checks and verify shared-deadline behavior with a deterministic clock/state test that rejects timeout-reset mutations. ## Verification Exact current head `e72e20ad111bd94728d38606015d1fdac3c92929` passed [full CI run 41142552195](https://github.com/paperclipai/paperclip/actions/runs/41142552195): all 49 workflow jobs reached accepted terminal results, including typecheck, build, every test lane and aggregate verify. All 59 reported checks are terminal: 55 success, three expected skips and one neutral 5/5 Paperclip review. Snyk passed. [Fresh Greptile review](https://github.com/paperclipai/paperclip/pull/18345#issuecomment-6528913235) explicitly names this exact head and gives 5/5. No review threads remain open. The PR is mergeable. These source gates do not establish a live QA pass at this head. Fresh protected QA at this exact source and the separately recorded deployed application/runtime remains required before primary merge and daily admission. Historical live QA staging **PASS** on qualified commit `af9e091b15b5e24b6b2cc1b424523430cc2ab024`: 6/6 checks, all five phases including cleanup, two native work runs, and the configured model/environment check passed. The [sanitized qualification report](https://pages.paperclip.ing/evals-qa-smoke-41122620146-1/) records that exact measurement. Final integration head `e72e20ad111bd94728d38606015d1fdac3c92929` merges exact actual squash `18b1637b6371a6a6adade857e2a73654ee626fef` from [#18254](https://github.com/paperclipai/paperclip/pull/18254) into previous head `e54d64743964b1a19d165537bebbe62c50389a23`. The merge had no conflicts. All 18 files in this PR's change set, the complete 45-file shared journey directory, and six shared native/evidence helpers retain their exact previous-head Git blobs: 59 scoped files in total. The production clarification helper remains blob `800a747284db19b8484650b6759e8bd1f356839d`. Follow-up 1–2 runs, question 1–3 runs, older exact counts, owned native event checks, active-before/after persistence proof, cleanup/accounting and feature mappings remain unchanged. Source integration head `e55d74491f5bcd99c1455e9ace02bd7ad7319fae` integrates exact actual squash `8595efc20bda3ce20a9fbbf34a5d156a6fcf6d57` from [#18385](https://github.com/paperclipai/paperclip/pull/18385). The merge had no conflicts. All 17 files in this PR's change set and six shared native/evidence helpers retain their exact Git blobs from previous head `162e341cda5b627639b486b4019a1675bb018ea2`. This preserves the follow-up and question readiness checks, feature mappings, six closed follow-up check IDs, immutable before/after active-operation proof, follow-up 1–2 runs, question 1–3 runs and older exact counts. Previously reviewed head `e54d64743964b1a19d165537bebbe62c50389a23` adds a separate test-only reliability repair on top of that source integration. It replaces one scheduler-sensitive browser case with a deterministic Vitest clock and locator-state model that calls the actual production helper. The six other real Chromium checks remain byte-for-byte unchanged and pass **6/6**. Focused unit suites pass **51/51**, including the deterministic deadline case. This is six browser checks plus focused unit coverage, not seven browser checks. The model makes the panel ready at 100 ms, the supplied answer ready at 600 ms, and full valid content ready at 1,300 ms. It requires rejection at the shared 1,000 ms deadline and records immutable rejection state before simulating a 10-second scheduler stall. Both restarted-timeout mutations fail: resetting all waits wrongly resolves at 1,300 ms; resetting matcher timeouts while keeping the final real deadline guard rejects only at 1,300 ms with valid content present. The genuine helper rejects at 1,000 ms with the answer still incomplete. The production helper is unchanged, Git blob `800a747284db19b8484650b6759e8bd1f356839d`. That two-file test-only patch changed no journey, application, run, accounting, model or runtime behavior. Its six-browser/51-unit results are prior-head evidence; the complete patch is unchanged in the new integration. The final source also retains the actual merged question-primary squash `d9354bbfc35ef74333e90e596ce21dc8e59149b9` from [#18346](https://github.com/paperclipai/paperclip/pull/18346). Later commits `409117451da260a03fc28a3107c7808ecbe68f8f` and `ea3e7b1d1bed562351fcf4722c7175001b987b59` bound question document readiness to one campaign deadline and test that waits cannot restart it. Commit `162e341cda5b627639b486b4019a1675bb018ea2` adds question/follow-up report feature mappings and coverage. These files are unchanged by the final merge. The imported upstream source has real application changes: [#18334](https://github.com/paperclipai/paperclip/pull/18334) makes managed skill content durable and extends native skill-test contracts; [#18350](https://github.com/paperclipai/paperclip/pull/18350) requires managed Cloud shared storage and bootstrap prerequisites. #18385 updates TestOps job metadata and repairs their combined maintenance fixture. The new merge additionally imports [#18383](https://github.com/paperclipai/paperclip/pull/18383) at `81ecaf0654fe79a2a2dbda6588f033e39ee610e4`, which bounds failed lifecycle verification; [#18357](https://github.com/paperclipai/paperclip/pull/18357) at `45a3a92e6dc52d20923b9aa34a22fee2ec5443f9`, which persists sandbox-creation ownership and updates cleanup/adapter/plugin contracts; and #18254 at `18b1637b6371a6a6adade857e2a73654ee626fef`, which accepts a valid late Antigravity authorization URL while native authentication drains. This PR's diff against exact upstream `18b1637b6371a6a6adade857e2a73654ee626fef` contains only journey tests, report coverage and documentation. Harness source and deployed application/runtime identity are separate inputs. This source integration does not deploy the imported application changes or claim the QA target runs this application revision. This final SHA has not been measured by the historical run above: **fresh QA at this exact final head is required before primary merge**, followed by normal TestOps verification/admission. Fresh exact-head CI and review have passed as recorded above. The live measurement must record the actual separately deployed QA application/runtime identity. Seven whole-file blobs remain identical to the qualified SHA. All paths are relative to `tests/runner-e2e/`: | Journey/helper | Unchanged Git blob | | --- | --- | | `product-journeys/active-followup-evidence.ts` | `83a5d9fab52abb9849a40e83eca6b6a9c29bf587` | | `product-journeys/active-followup.test.ts` | `728160c610c18292d2cf06d4a87d25a8719f1734` | | `native-task-reply.ts` | `59d11f1a178dad9db3da8fa2342ff93643cb9922` | | `native-event-envelope.ts` | `305b0b456c59c511ee328e529ad574c0583075a0` | | `run-observations.ts` | `08c8ad184a821a7cd536926b3d7e3d1156a1e3da` | | `user-actions.ts` | `e9611348dc41e17899367dab2e3b36d3d1be9480` | | `stop-redirect-flow.ts` | `4580bd605a425bc72511e02d988492acbec8d60e` | The follow-up preservation comparison against historical `af9` has three explicit exceptions; no whole-file identity is claimed for them. The separately documented question readiness and report additions above also require final-source verification: 1. `product-journeys/agent-flows.ts` adds `question-answer-resume` to the `AgentJourney` type exclusion, alongside follow-up. Its remaining contents and runtime helper bodies are byte-for-byte unchanged. Qualified blob `a0afaa968c1d854476159fef21de7b6dd50a4a98`; final blob `0cc797c0ae594e7769338e63a80c4f67e29e4a54`. 2. The common run-count helper is centralized in `success-contract.ts` and re-exported by the catalog. It accepts follow-up **1–2**, clarification **1–3**, and each older paid journey's exact count. The follow-up catalog entry, maximum of two runs, six required checks, native dispatch, collection and retained-run contract remain byte-for-byte unchanged. The helper's text differs; its preserved count semantics are verified explicitly. 3. `product-journeys/active-followup-flow.ts` changes exactly one composer readiness assertion from Playwright's implicit five-second default to an explicit 30-second timeout, matching `submitTaskReply` for attached targets. Prompts, actions, native evidence, before/after persistence proof and grading otherwise compare byte-for-byte unchanged. Qualified blob `64392329198a1a4d9828e32834d4507e175216e6`; final blob `2746b5ba89e2af26b4097ba7897913186786ac5d`. This correction is why a fresh final-head QA run is required. - **310 focused tests passed at current final head `e72e20ad111bd94728d38606015d1fdac3c92929`**: 286 journey/evidence/catalog/report/shared-deadline tests in 14 files and 24 native authentication/real-child stream-order tests in two files. Both invocations used Node24 and one Vitest worker. The earlier `e55` integration also passed 365 focused tests, including 40 Cloud configuration and 40 TestOps metadata cases; those results remain prior-source evidence. - **157 explicit count assertions passed again at the current final head**, including both interactive ranges, older exact counts, invalid/fractional/non-number values and catalog/contract export identity. The maximum bounds and six closed follow-up check IDs also matched. - Maps and diff checks passed. The official QA runtime-import preflight passed on Node24 at the current final head without network access. It uses a dummy adapter and no credentials. It does not prove target connectivity, sign-in, model readiness or deployed application/runtime identity. Those require the normal protected QA run. No model/provider work or optional Runner release was dispatched. - Current local runtime dependencies use ignored isolated workspace-source views. No shared dependencies or compilers were changed. A previous local typecheck attempt hit stale reused Runner declarations. No full repository typecheck, test run or build was repeated locally; the successful exact-head full CI run above now supplies these gates. - CI on previous head `162e341cda5b627639b486b4019a1675bb018ea2` failed the Cloud/local maintenance fixture in [server shard 9](https://github.com/paperclipai/paperclip/actions/runs/41133716802/job/124331643100), plus dependent verify. The actual tested merge combined that head with master `198218bf554877fe3b7a83742eefb3b8f12392c6`. A Cloud child selected local-disk storage and hit the new shared-storage policy. This was a deterministic upstream fixture mismatch; no unchanged failed-job retry was used. - The actual #18385 repair configures a valid Cloud child, verifies local bytes independently, checks the expected provider mismatch and retains exact stored-config readback. Its own final head `b76d5dd40d7711f1b5f6b6316b1114c0aae02d57` passed 59 checks, including [server shard 9](https://github.com/paperclipai/paperclip/actions/runs/41133876636/job/124331450635). That prior result supports the repair; it does not replace this final head's fresh CI. Separate repair candidate #18392 and its source/evidence are retained as superseded history. - A local attempt to run the database fixture could not start PostgreSQL because the host exhausted macOS semaphores. A first setup attempt also failed package resolution before test execution. Skipped database cases and these setup failures are not passes. No kernel settings, shared processes or databases were changed. - The complete previous-head CI run [41138618869](https://github.com/paperclipai/paperclip/actions/runs/41138618869) finished with one underlying failure: [Runner Vitest shard 1](https://github.com/paperclipai/paperclip/actions/runs/41138618869/job/124347859445) expected native Antigravity login success but observed failure at `antigravity-auth.test.ts:101`. The lane had one failed test, 1,994 passing tests and eight skipped tests. Aggregate verify failed only because of this lane. All other 57 checks were accepted, including build, typecheck, all server/browser suites, Runner shard 2, Snyk and exact-head 5/5 reviews. No unchanged retry was requested. - The independently reviewed #18254 repair removes an invalid cross-pipe ordering assumption: a valid URL on stderr can arrive after the authenticate response on stdout during drain. Its real-child regression forces both orders. The current final source imports that actual squash; the 24 focused authentication cases pass here. Fresh full CI and exact-head review have now passed for this source, as recorded above. The earlier failed result is not relabeled. - The original live compatibility failure is retained: an older PRP v1 pack labeled builtin `bash` as operation `unknown`, causing the observer to miss the command and never send the follow-up. Its exact shape was reproduced red before repair. Generic unknown operations, semantic tools, wrappers, read-only work and foreign provenance remain rejected. The historical qualified pass used the repaired oracle; no failed result was relabeled. - Earlier CI on `6fad965d69` hit an unchanged heartbeat test's three-second adapter-call wait; its original failure and one narrow rerun remain recorded. The qualified SHA subsequently passed CI. An earlier report-catalog subprocess failure is retained. The final focused Node24 run now includes that suite and passes after the documented report feature mappings. ## Risks The fixture asks the native agent to run one bounded `sleep 90` command. The observer must see actual active native execution; a fast or noncompliant turn fails rather than passing from elapsed time. A completed result requires one or two bounded native work runs, source attribution, configured model/environment, and collection. Unknown provider prices stay unknown. This change does not alter production execution or grant new access. ## Model Used OpenAI Codex, GPT-6 family, with reasoning and code execution. The host does not expose an exact underlying model version or context-window size. ## Checklist - [x] I have included a thinking path that traces from project context to this change - [x] I have specified the model used (with available 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; existing follow-up execution fixes do not provide this deployment-targeted test - [x] I have described the issue in-PR following the enhancement template - [x] I have not referenced internal/instance-local Paperclip issues or links - [x] My branch name describes the change and contains no internal ticket or instance-derived details - [x] I have run the new focused tests locally and they pass - [x] I have added or updated tests where applicable - [x] I have updated relevant documentation - [x] I have considered and documented risks - [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 merge --------- Co-authored-by: Paperclip <noreply@paperclip.ing>
12 KiB
navigation
| navigation | ||||||
|---|---|---|---|---|---|---|
|
Steering and queued messages
People can send follow-ups while an agent works, manage saved messages that have not been consumed, and choose when to steer or interrupt. Stopping a run and pausing a task are separate controls. The UI must preserve the message, explain why it is waiting, and only claim delivery after an authoritative acknowledgement.
Feature ID: steering.
Agent access: companion.
Sub-features
queue:steering/queue— a busy task saves follow-ups and shows their order and wait reason.edit-order-discard:steering/edit-order-discard— edit full Markdown, reorder, and discard eligible saved messages; ownership, revision conflicts, and already-dispatched states matter.steer:steering/steer— explicit user action sends a queued message into a compatible active run; unavailable capabilities and provider rejection remain visible.interrupt:steering/interrupt— stop the current execution and carry accepted follow-ups into the eligible continuation without delivering them twice.stop-and-pause:steering/stop-and-pause— distinguish stop-in-flight, confirmed stop, task/ancestor pause, agent pause, and budget hold. Resume must respect the actual gate.drafts:steering/drafts— switching tasks, reloading, and uncertain submission retain the correct task's draft and attachment receipts.approval-queue:steering/approval-queue— answered questions and accepted approvals have durable delivery distinct from editable ordinary comments.
How to get to it (user POV)
task-composer
Open an assigned task (/issues/:issueId). Start work, then send follow-ups while
the agent runs. Use the queued-message controls to edit, reorder, discard,
steer, or interrupt when offered. After Stop, choose Send on a saved queue to explicitly continue it without recreating the messages or attachments. Native steering still requires an active turn; Send after Stop is a separate user decision. Use the composer stop/pause control separately.
Both task presentations need attention when changing their shared behavior.
agent-chat
Enable Agent Chat and open an agent conversation. Send another message during
a reply and inspect the shared queue. /new starts fresh provider context in
the same conversation; it is an ordered boundary, not a normal model prompt.
request-response
Answer an earlier question or accept a confirmation while successor work is active on the task. Inspect the resulting queued response and its available steer/interrupt control instead of assuming acceptance interrupted execution.
external-channel
Send a follow-up from a linked external chat conversation while its task is running. Use that provider's supported stop/control affordance where available; inspect the resulting task and run in Paperclip.
Driving it
The explicit stop-redirect Product journey covers steering/stop-and-pause only for Stop and one new direction. It requires active work, native cancellation acknowledgement, resource cleanup, and completion without later original-turn activity. It does not qualify pause, interrupt, or all queued-message behavior. Source registration is not a live Cloud pass.
The explicit followup-while-running Product journey
covers steering/queue: one follow-up is saved while a bound native command is
active, its late requirement appears in one attributed final reply, the queue
empties, and the result survives reload. One steered turn or one normal successor
is allowed. Unit calibration and source registration are not live acceptance.
Preconditions: follow the baseline. Use an agent that stays busy long enough to inspect the queue. Record the adapter, native/legacy mode, and advertised steering capability. Test supported and unsupported steering without treating an absent capability as a product failure.
task-composer
Automated: queued-message UI covers order, failure restoration, unavailable steering, stale revisions, and discard acknowledgements. Queue routes cover durable ordering, authorization, steering acknowledgements, interrupt recovery, and concurrent dispatch. Composer covers drafts and uncertain saves.
Task-detail delivery races cover a consumed message whose send response and queue snapshot arrive late. A started run's input receipt removes that message from pending, while a separate unsent message stays queued. Admission to a run that has not started is not delivery. A later retry of the same message remains visible in its new authoritative queue. A queued turn's own starting input stays in history before execution, including board messages that reopen a task. Later inputs keep their queue controls and full canonical order. A maintenance hold displays its current admission wait; drain release and expiry prompt normal admission without a manual retry. Receipts retain the primary wake and coalesced source queue IDs. They clear a late snapshot even when its target names the preceding run or is null.
pnpm exec vitest run ui/src/components/task-chat/TaskChatQueuedMessages.test.tsx server/src/__tests__/issue-queued-comments-routes.test.ts
Automated stop coverage is separate: composer stop with its configuration and ACP stop/continuation. Inspect their fixture and opt-in prerequisites; skipped native cases are not native proof. The explicit follow-up/Stop/resume journey requires a later user message before continuation. The database Stop accounting regression checks complete and late receipts without another provider turn. These are separate scopes; database success does not establish deployed or Slack acceptance.
Manual: queue three distinguishable messages. Edit one, reorder them, reload, and confirm the saved order. Discard one and confirm it never reaches execution. After a queued message enters a started run, confirm it leaves pending even if the send response arrives later. Reload and confirm it appears once in history. Steer one while busy and verify the agent receives that exact message once. Repeat with a rejected steer and a stale second-tab edit; the UI must preserve retryable content and explain the conflict. On a fresh run, interrupt with a queued follow-up and verify confirmed stop followed by one continuation. Pause the task (and separately its ancestor), preserve a draft, then resume the actual hold; no hidden send should occur while paused. Check both task presentations when available; the named component tests do not prove every host integration.
Fixture-backed recovery check: start with an agent in its completed paused
lifecycle state, then insert a new accepted deferred_issue_execution request
through the PostgreSQL recovery fixture.
Normal wake admission
rejects paused agents, so this pending state cannot be created by sending a new
message through the ordinary UI or API. Let release and periodic recovery run.
Confirm that the saved request remains pending. Resume the agent, wait for its
lifecycle to become ready, and verify one delivery. An explicit Pause cancels
older queued work; those cancelled receipts must not be replayed. A paused
request must not prevent another eligible recipient's queued work from proceeding.
agent-chat
Automated: agent chat sessions covers persistent conversation/session behavior with deterministic harness fixtures; shared composer tests cover the controls. These do not qualify live steering for every harness.
Manual: queue two follow-ups while the conversation replies. Reload, edit or
discard one, then let the other run. Verify one reply and unchanged conversation
identity. Send /new, then a new prompt; old history remains visible, and the
new turn uses fresh provider context. Verify a useful answer rather than only
the session-divider UI. Repeat the queue operations with a live capable harness
when the change touches delivery.
request-response
Automated: response delivery proves queueing/coalescing and retry-safe receipts. The queue-route suite tests separate comment and approval queues, explicit steering, and fresh-session restrictions on plan approvals.
Manual: answer an old question during a new run, confirm that it is saved and queued, and let the active run end. Verify exactly one continuation with the answer. Repeat an eligible approval with explicit steer; another approval that requires a fresh session must not offer same-turn delivery. Do not edit an approval outcome through an ordinary queued-message editor.
external-channel
Automated: chat interaction arbitration covers backend arbitration with simulated requests. It is not a provider UI test.
Manual: send two distinguishable messages through one disposable linked conversation while busy. Inspect their saved task order, run delivery, and external reply. Check a duplicate provider event and unauthorized sender. Provider-specific control semantics and real transport timing remain a manual gap; do not assume browser queue controls are available in every channel.
Gotchas
- Paperclip's queue is server-owned once saved; it is not Omnigent's unsent browser draft buffer. A stale edit must not overwrite a consumed message.
- Steer availability depends on the active runtime's capability, not its brand name. A UI row disappearing optimistically does not prove provider receipt.
- Interrupt, stop, task pause, agent pause, and budget pause have different effects. A running indicator disappearing does not prove remote work stopped.
- An interrupted run may need verified cleanup before a saved follow-up can start. Preserve and inspect the wait reason; see recovery.
- A healthy idle Agent Chat is waiting for input, not stranded unfinished work.
Storybook verification
drafts
Capability ID: steering/drafts.
Scope: ui Storybook: Task-keyed restored draft and attachment receipts; task switch/reload, failed send and uncertain-save recovery. Production: Retain correct task draft across navigation and reload; inspect authoritative saved receipt before claiming delivery; do not clear storage on every harness mount. Sources: Source: TaskChatComposer.tsx, composer-draft.ts. Tests: TaskChatComposer.test.tsx, issue-queued-comments-routes.test.ts
approval-queue
Capability ID: steering/approval-queue.
Scope: ui Storybook: Answered question/accepted approval delivery waiting, eligible steer/interrupt, failed delivery and fresh-session-only approval. Production: Keep response delivery durable and distinct from editable comments; coalesce once, enforce fresh-session restrictions and deduplicate concurrent/retried continuation. Sources: Source: TaskChatQueuedMessages.tsx, question-response-delivery.ts. Tests: TaskChatQueuedMessages.test.tsx, question-response-delivery.test.ts, issue-queued-comments-routes.test.ts