Nicky LeachandPaperclip
35ab90ad0f
## Thinking Path > - Paperclip manages agents and tasks. > - Plugins contribute views to the task sidebar. > - A ready view can belong to an earlier task. > - Runner records activity under its authenticated run. > - This PR maps that activity to installed task-panel declarations. > - The sidebar checks readiness after current-run use and restores use after reconnect. ## Linked Issues or Issue Description Builds on merged contract #18129 and Runner publisher #18050. This PR targets master and is ready for review. Earlier combined verification is retained in [the verification archive](https://github.com/paperclipai/paperclip/pull/18050#issuecomment-6489369676). **Subsystem affected** Run-event reads, plugin task contributions and the task sidebar. **Problem or motivation** Readiness polling can show a view left ready by another task. Reload and reconnect must recover current-run use without losing concurrent live activity. **Proposed solution** Map recorded activity to installed task-panel declarations. Merge authorized history and live activity, then perform bounded readiness checks for the running agent. Preserve the existing read-only view. ## What Changed - Add optional `{ namespace, activityKey }` activation mappings to task-panel declarations. Legacy declarations remain valid but contribute no task panel. - Add `view=plugin-panels` to the existing run-events route. Preserve run-read and telemetry-read authorization, secret redaction and private-run denial. Map activity to installed contributions and return at most 128 panel identities. - Scope activation caches to current panel versions and activity declarations. Preserve unchanged panels during unrelated plugin updates and history reads. - Merge history and live hints without losing concurrent activations. Filter by company, task, run and agent. Restore history on mount and reconnect. - Replace periodic availability checks with up to three readiness attempts after activity. Fresh activity restarts settled transient failures. Pending checks, successful readiness and HTTP 4xx failures are left alone. - Reuse live run updates with the panel cache key. Recover only panel history after a pending subscription read. Mark expected not-ready query results so the global error reporter skips them; real failures remain reportable. - Preserve current-run dismissal and the read-only plugin view. Update manifest comments and producer/consumer documentation. ## Verification - Current head: `5e04d757c`, based on master `09c05c550`. Range comparison confirmed all six feature commits were unchanged by the rebase. The final follow-up keys readiness by the plugin installation ID and availability path. - `cd ui && node ../node_modules/vitest/vitest.mjs run src/hooks/useTaskPluginPanels.test.tsx` — all 45 cases passed. `cd ui && node ../node_modules/typescript/bin/tsc -b` passed. - The shared query-client and live-update recovery selection passed 96 cases on `49c544cd2`. Token gates passed before the final hook-only changes. - Earlier focused verification passed 57 manifest cases, 69 database/history cases, event-registry cases and route authorization cases, including private-run denial. - Regression coverage includes subscription/history races, task changes during recovery, unfocused run completion, bounded readiness retries, denied reads, expected not-ready responses, changed activity mappings, unchanged panel preservation, old cache scopes, and same-version route or installation changes. - [Current-head CI](https://github.com/paperclipai/paperclip/actions/runs/40942001707) passed: 54 successful checks, two skips, no failures or pending checks. - [Standard and final review](https://github.com/paperclipai/paperclip/pull/18131#issuecomment-6489842168) both scored 5/5 on `5e04d757c`. The final review used the Apex tier, added no comments, and left no unresolved findings. The PR is conflict-free and ready for review. - Local checks stay targeted. GitHub CI provides the broad build, test, typecheck, browser and Runner coverage. No deployment or live-provider acceptance is claimed. <details> <summary>Retained failures and recovery evidence</summary> - `ac9329548` twice failed a browser reload assertion while conversation history was concealed. Master fixed its readiness wait in #17923. The approve journey then passed locally on `fea63a418` with installed Chrome (1 passed, 39 seconds). All eight browser shards passed in later CI. The browser test is outside this diff. - A canary checkout failed with Git HTTP 408; its unchanged-head retry passed. - `de07d09fa` failed the mobile Cases card click after a ten-second stability wait. The same read-only journey passed locally on that exact head (1 passed, 26 seconds). Its unchanged-head CI retry passed 22 cases, with one skip. The original artifact is retained; the cause of that click timeout is not established. - [CI on `49c544cd2`](https://github.com/paperclipai/paperclip/actions/runs/40903679075) passed 54 checks with two skips. CI also passed on `de07d09fa` after the recorded retry. - The `1e59ab441` merge checkout passed all browser and server suites. Runner shard 2 passed 2,252 cases and timed out waiting for a controller-recovery command. The test had landed on master in #18161 after the former branch base. Rebasing aligned the local checkout with that code; Runner shard 2 then passed on `fcf7380d3`. The original timeout cause remains unproven; no unrelated timeout was changed. - [CI on `fcf7380d3`](https://github.com/paperclipai/paperclip/actions/runs/40940235193) recorded runner shutdown signals in typecheck and a workspace shard. A chat recovery test observed a null wake deadline, and one Runner transport fixture failed during close. Those separate assertion failures remain unexplained. All of those jobs passed on the final readiness-key head. Those later passes establish recovery; they do not prove the cause of the earlier assertion failures. - Local browser setup first stopped at pnpm version-download verification, then a sandbox IPC restriction, then a missing bundled browser. Installed pnpm 9.15.4 and Chrome completed the journeys. No dependency files were changed. - Earlier optional-property TypeScript errors were corrected. Embedded PostgreSQL needed its install hook run from its package directory. Older CI also recorded fixture-cleanup lock timeouts, attention-fixture timeouts, and canceled server shards after runner shutdown. The parent fixed its FIFO snapshot-copy failure before merge. The linked verification archive retains earlier combined evidence. </details> ## Risks Activity is a hint and grants no access. Missing plugins, denied reads and unavailable views contribute no panel. Older runs without activity records show no task plugin tab. A used view stays available for the running task; reconnect reconciles readiness. Matching producer and consumer builds are required. No migration, dependency, image publication or deployment is included. ## Model Used OpenAI Codex, based on GPT-6, with reasoning, repository tools and code execution. The exact serving snapshot is not exposed. ## 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>
4.9 KiB
navigation
| navigation | ||||||
|---|---|---|---|---|---|---|
|
Delegation, dependencies, and signoff
A larger task can coordinate child work, dependencies, reviewers, and execution controls while preserving a clear owner and a traceable completion decision.
Feature ID: task-coordination.
Implementation: task thread, execution policy.
Agent access: companion.
Sub-features
delegation:task-coordination/delegation— inspect parent/child tasks and the original request behind delegated work.dependencies:task-coordination/dependencies— show blocking tasks and wake eligible work when prerequisites resolve.signoff:task-coordination/signoff— apply the configured implementation/review completion policy.tree-control:task-coordination/tree-control— pause or resume a task tree without bypassing its gates.
How to get to it (user POV)
task-relationships
Open the task’s parent/child and dependency links in task detail.
review-and-controls
Inspect the task’s execution policy and available review, pause, resume, or stop controls.
Driving it with Playwright and the Paperclip API
Preconditions: follow the baseline. Prepare a parent, child, and dependent task plus implementer/reviewer agents. Record execution policy and use a disposable tree.
Parent/dependency navigation and inherited tree holds can be checked on unassigned tasks. Agents are required for the continuation and signoff portion. A tree pause creates a hold; it does not change task statuses to a paused status.
task-relationships
Automated: dependency wakeups checks server behavior; visual relationship navigation remains a manual check.
Manual: Block one task on another, finish the prerequisite, and observe the dependent task’s eligible continuation. Follow parent/child links back to the source request. Verify no duplicate run or false parent completion occurs.
review-and-controls
Automated: execution policy, tree control, and signoff browser suite target these layers. A linked suite is not a claim that its current run is green.
Manual: Complete implementation under a review-required policy and confirm it waits for the required review outcome. Exercise rejection and later approval. Pause the parent while child work is pending, then verify resumption respects budget and ownership gates.
Resume removes the pause without changing task status. Its wake count includes only current agent assignees on todo, in-progress, or in-review tasks. The dialog explains that blocked, backlog, and unassigned tasks remain idle. Verify this in both task interfaces with the paused composer browser suite. Release rechecks tasks reopened after the pause; another active pause still prevents their execution, and uncertain prior executions must be reconciled before waking.
Plugin task panels
An installed plugin can offer an agent detail tab in the task side panel through
its optional taskSidePanel declaration. The host supplies the task's running
agent and run. The runtime captures Runner context before use and records
runtime.activity.recorded after success. Runner rejects stale contexts and knows
no plugin or view identity. Core maps the recorded namespace/activity key to the
panel's declared activation pair and restores it from the run log on reload or
reconnect. Readiness gets three bounded attempts after matching activity, with
no periodic availability poll. Declarations without a mapping add no task panel.
The plugin's authorized availability route decides whether the panel is
shown, and the plugin owns its label and viewer. No contribution appears for a
missing plugin, inactive run, unavailable view or access failure. A closed panel
can be reopened from the launcher while available.
Implementation: panel discovery and task panel. Automated: discovery tests and task panel tests cover availability, task/agent changes, failures, inactivity and dismissal. These mocks do not establish a particular plugin's live transport behavior.
Gotchas
- Review completion and task completion are different events under some policies.
- Pausing a tree must not be “tested” by changing statuses until the screen looks quiet.
- Use questions and approvals for decision forms and recovery for interrupted work.