Devin FoleyandPaperclip
6c198d616e
## Thinking Path > - Paperclip is the open source app people use to manage AI agents for work. > - Tasks have one owner, and the server records assignment changes. > - An agent can finish its work by handing the task to a human for review. > - The assignment route stops the active run before it saves that handoff. > - When the caller is that run, a slow workspace restore can exceed the stop timeout and prevent the assignment from saving. > - Recovery can then mark the task Blocked while its comment says the handoff succeeded. > - This pull request lets the authenticated legacy run save a human handoff and finish its turn normally. > - Other actors and transfers to executable agents keep the verified stop requirement. ## Linked Issues or Issue Description Refs #13730. Related: #13782, #14700, and #15094. **What happened?** An agent posted its review handoff, then sent `PATCH /api/issues/{id}` to assign the task to its human creator with status `in_review`. The route cancelled that same run before saving the assignment. Workspace restore took 66 seconds. The stop limit is 60 seconds. The assignment never committed. Recovery marked the task Blocked, and it stayed assigned to the agent. **Expected behavior** The authenticated active legacy run can hand its own task to a human and finish normally. The human assignment and requested review status must save without waiting for workspace restore. Native runs, board reassignment, agent API keys, and transfers to another agent still require a verified stop. **Steps to reproduce** 1. Run a legacy agent on its assigned task with a signed run token. 2. Send a task PATCH from that run with a human `assigneeUserId`, a null `assigneeAgentId`, and status `in_review`. 3. Make stopping the sandbox take more than 60 seconds. Before this fix, the handoff fails before the task update. With this fix, the update saves and the run finishes normally. **Paperclip version or commit** Reproduced in hosted execution. The patch is based on master at `56a0bf0d492`. **Deployment mode** Hosted authenticated server with a legacy agent in a remote sandbox. The same route serves local deployments. PR #13782 proposes a broader exemption for all self-handoffs. This focused change limits the exemption to human recipients and legacy runs. It checks the exact active run, agent, company, and task, and it retains the stop requirement before an executable successor can take ownership. PRs #14700 and #15094 address later wake and recovery behavior after a reassignment stop. ## What Changed - Skip the reassignment cancellation only for an authenticated legacy run handing its own task to a human. - Reject later mutations from the outgoing run against its human-owned task, both in route authorization and under the update transaction locks. - Keep assignment permission checks, goal control, and queued human response protection. - Add route regressions for slow stops, active-run lookup, actor and runtime boundaries, recipient types, permission denials, and follow-up edits, completion, and reassignment. - Update the real HTTP low-trust handoff test to prove the run stays alive, its late edit fails, and the human wake receipt remains invalidated. - Add Postgres tests for saved assignment, former-owner rejection, and review state after the outgoing run succeeds or fails. - Document the handoff rule and update the task feature and agent access maps. ## Verification - 328 relevant tests passed across the route, issue service, low-trust HTTP, and legacy recovery suites. The route suite passed all 160 tests, and all 13 low-trust HTTP tests passed after updating the obsolete cancellation expectation. - Removing the production guard makes both slow-shutdown handoff regressions fail with HTTP 500. - `pnpm -r typecheck` passed. - `pnpm build` passed. - `node scripts/check-maps.mjs` passed. - `git diff --check` passed. - All 54 current-head checks passed; 2 optional Storybook checks were skipped. The CI lanes cover the full Vitest suite, typecheck/build, Runner checks, and product E2E checks. - Apex reviewed `c68fbfb356c3d80eb7136babb49940709a3c3459` at 5/5 with no findings; both earlier review threads are resolved. - The complete test lanes passed in CI. The duplicate local full `pnpm test:run` was stopped after CI completed; it did not finish locally. All 328 focused local tests and full local typecheck/build passed. ## Risks - The outgoing legacy turn stays alive after the human takes ownership. Route and transactional ownership checks reject its later mutations of the human-owned task. Read and comment access follows normal policy. Goal control still runs before the handoff. - This exemption does not apply to native runs or another agent as recipient. Those paths still stop the old execution before assignment. - This fix prevents future failed handoffs. It does not infer or replay old assignment requests. - There is no schema change or API shape change. ## Model Used OpenAI GPT-6 through Codex, with reasoning, repository tools, and code execution. The session does not expose a more specific runtime model ID 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 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>
Paperclip feature map
Start here when reproducing a user-facing bug, verifying a change, or deciding which surfaces a fix must cover. Each recipe describes what a person can do, where they can do it, the expected result, and the evidence needed to verify it. Product behavior is still governed by the implementation spec.
The sibling agent access map records APIs, agent entry points and permission boundaries for the same capability IDs. Follow the shared map contract when maintaining either map.
Features
The map is organized by user capability, not by source file. It covers 35 feature families across the product, CLI, and operator workflows, checked against source on 2026-10-12. Experimental and developer-only surfaces are labeled explicitly. Each recipe includes sub-features, current entry points, automated evidence, manual verification steps, and gotchas.
Getting started and access
| Feature | What it covers |
|---|---|
| Onboarding and first work | Instance setup, company wizard, first agent and task. |
| Login, invitations, and access | Sessions, bootstrap, membership, roles, and CLI authorization. |
| Companies and portability | Company switching/settings, archival, package import/export. |
Tasks and conversations
| Feature | What it covers |
|---|---|
| Task creation and lifecycle | Creation, assignment, lists, properties, comments, and completion. |
| Delegation, dependencies, and signoff | Child work, prerequisites, reviewers, and task-tree controls. |
| Inbox, decisions, and search | Personal triage, decision queues, unread/blocked views, and search. |
| Agent conversations and project handoff | Persistent conversations, discovery, handoff, and gated board chat. |
| Questions and approvals | Question/plan responses, formal approvals, and app-tool review. |
| Steering and queued messages | Follow-ups, queue edits, interruption, and pause/resume. |
| Documents, attachments, and work products | Versioned documents, annotations, files, outputs, and artifact library. |
Agents and reusable capabilities
| Feature | What it covers |
|---|---|
| Hiring, configuration, and organization | Agent identity, instructions, reporting lines, lifecycle, and built-ins. |
| Runs, harnesses, and model accounts | Adapter/model setup, account validation, transcripts, and run history. |
| Skills and Skill Studio | Discovery, sources, authoring, revisions, tests, and agent policies. |
| Team packages and installation | Catalog preview/install via CLI/API; catalog UI availability called out. |
Projects and execution
| Feature | What it covers |
|---|---|
| Projects and repositories | Project lifecycle, task context, repository configuration, and defaults. |
| Goals and work alignment | Goal hierarchy, ownership, status, and project/task context. |
| Workspaces, services, and files | Provisioning, task bindings, services/logs, Git/files, and closure. |
| Execution environments | Local, SSH, sandbox providers, target probes, and custom images. |
| Recovery | Stopped work, workspace repair, bounded continuation, and reconnect. |
Apps, channels, and extensions
| Feature | What it covers |
|---|---|
| Connection setup | Catalog, MCP links, task requests, authentication, and setup resumption. |
| App access and action permissions | Agent grants, off/ask/allowed actions, discovery, tests, and revocation. |
| External chat and email | Provider-specific identity, threads, files, delivery, and endpoint upkeep. |
| Tool gateways and access profiles | Tool exposure, client configuration, tokens, profiles, and activity. |
| Secrets and proposals | Company/user credentials, grants, proposals, and vault import. |
| Plugins | Installation/configuration, contributed pages/tools, and lifecycle. |
Automation and structured work
| Feature | What it covers |
|---|---|
| Routines, schedules, and triggers | Definitions, variables, scheduled/webhook/manual runs, and history. |
| Pipelines, review queues, and learnings | Experimental stages, automation, items, review, and learning records. |
| Cases | Experimental structured fields, relationships, revisions, and task links. |
| Status cards | Experimental summaries, watched work, refresh history, and settings. |
Oversight and operation
| Feature | What it covers |
|---|---|
| Costs and budgets | Spend reports, scoped limits, incidents, hard stops, and resumption. |
| Dashboards and audit trails | Company health, live work, activity, runs, routines, and timeline. |
| Navigation, profile, and announcements | Sidebar state, favorites/recents, personal identity, and dismissals. |
| Instance operations | Installation, updates, service health, configuration, and backups. |
| CLI/API and local worktrees | Explicit context, resource commands, outputs, runs, and isolated instances. |
Contributor surfaces
| Feature | What it covers |
|---|---|
| Developer previews and diagnostic labs | Design examples, interaction fixtures, performance checks; not production acceptance. |
These are verification instructions, not a claim that every journey passed a live test. A component test proves its component; a scripted provider proves that fixture integration. Record executed results separately from the map.
Before driving a journey
- Use this checkout's isolated test drive for manual product checks. It creates a temporary instance and prints its URL and data directory. Do not guess that a server on port 3100 belongs to you.
- Record the commit, URL, company, login role, relevant experimental settings, adapter, runtime mode, and live versus simulated dependencies. Use disposable tasks and provider resources. A fresh test drive creates a company and CEO, but no task or first run.
- Use the current navigation and company prefix. Paths in recipes omit that prefix. Record whether the streamlined or production shell is selected; Agent Chat, chat connectors, and the combined Inbox/Tasks view have separate gates. A hidden surface is an unmet prerequisite, not a successful test.
- Run the smallest relevant test first. Vitest commands below run from the repository root after installing dependencies. Playwright recipes use their named configuration and its isolated environment, not an unrelated running instance. Reserve expensive runner/provider tests for the behavior at issue.
- Exercise the actual user action, inspect its result, reload, and verify the persisted state or continuation. Use read-only API checks to corroborate UI evidence; creating state through the API does not prove the creation UI.
For agent-driven acceptance work, start with the Paperclip verification skill for an isolated launch, ownership/readiness checks, browser driving, evidence, and cleanup. It uses this directory as its feature map. The existing dev-workspace run/verify skill and evaluation skill describe runtime ownership and evidence handling. Reuse them; this map introduces no environment launcher, credentials, scheduled job, or second test framework.
Evidence contract
Report each entry point as passed, failed, blocked (with its missing prerequisite), or not run. Include the feature filename and entry-point ID, commit/environment, user action, expected and observed result, command/exit code, and evidence links. Screenshots or traces should show both the action and the discriminating result. Include task/run/request IDs when relevant, without secrets.
A fix is verified across its affected surfaces only when each has evidence or an explicit reason it does not apply. Shared code alone does not prove parity. Keep product regressions visible; do not change the recipe to bless a failure. For Paperclip-assigned work, attach evidence through the artifact workflow.
Inventory and remaining depth
The UI coverage inventory accounts for every non-test TSX
module under ui/src/pages, including supporting panels, legacy variants, and
labs. Every product area now links to concrete recipes. Partial means the
area still has the named variant or workflow gaps; unmapped is reserved for
an area with no recipe. These statuses describe documentation, not runtime health.
The page inventory is a source snapshot dated 2026-10-05. The map also includes entry points hosted inside other modules (pipeline Review Queue and Learnings, task documents, onboarding), CLI-only operations, and operator work. The team recipe explicitly distinguishes the current CLI/API path from catalog UI components without a current top-level route. Refer to route registration and the CLI registry when changing reachability.
Remaining depth includes complete provider/auth/attachment matrices, every harness/model/environment capability combination, third-party plugin features, and every role/error/mobile/legacy-shell permutation. The recipes name relevant gaps instead of equating source presence with a working user journey. The index can include capabilities that do not add a page file.
Maintaining the reference
The map documents user journeys. It does not execute them automatically;
coverage.json records documentation scope rather than live test health.
When maintaining a recipe, compare its entry points and expected results with
the current source. Check that linked tests still exercise the stated behavior
and that local references still exist. Refresh the inventory snapshot when it
helps explain the current product. Keep the same-named agent access companion
and stable sub-feature IDs aligned when behavior or access changes. Source-only
access inventory remains distinct from live verification. Run
node scripts/check-maps.mjs after map edits for structural/reference checks;
follow the shared contract for the source and behavioral checks it cannot perform.
Each recipe starts with an H1 and a user-visible description, then exactly:
- Sub-features — stable backticked IDs and observable behavior/states.
- How to get to it (user POV) — one H3 backticked entry-point ID per surface.
- Driving it with — starts with
Preconditions:; repeat each entry-point H3, withAutomated:andManual:paragraphs. Name test scope and gaps honestly. - Gotchas — misleading look-alikes, feature gates, and invalid evidence.
New or updated entries follow
Poteto's feature-map examples.
Existing entries with a Driving it heading can adopt the named harness when
their recipes are updated. The original format was inspired by
Omnigent's feature map.
Paperclip's recipes and checks follow its own task model and test infrastructure.