## Thinking Path
> - Paperclip manages work for AI agents.
> - Native helpers read task documents through authorized API tools.
> - A scan of 146 cases needs one saved response and one download per
case.
> - The document service can read the same key across authorized tasks
with bounded queries.
> - This pull request adds a bounded bulk read with the existing task
access policy.
> - Helpers can transfer one batch and check current revisions before
fetching changed bodies.
## Linked Issues or Issue Description
**Subsystem affected**
Task document REST API, shared query contracts, and native API
discovery.
**Problem or motivation**
Per-task document reads make native sync scans cost hundreds of tool
calls. Task list timestamps do not prove that document content is
unchanged.
**Proposed solution**
Add `GET
/api/companies/{companyId}/issue-documents?key=notes&issueIds=...`. Read
up to 200 task UUIDs under current company and task permissions. Return
current bodies and revision metadata, or metadata alone with
`includeBody=false`.
**Alternatives considered**
Generic request batching would add tool execution and partial failure
rules. One bulk resource read removes the per-case relay cost with the
existing tools. Do not skip documents based only on task-row timestamps.
Keep managed skill files immutable; cleanup of copied skill test
fixtures belongs to those tests.
**Roadmap alignment**
This improves existing task documents and native API access. It adds no
new workflow or UI. The roadmap has no duplicate bulk document read
work. No duplicate open pull request was found.
## What Changed
- Add a bulk keyed document route and a company-scoped document reader.
Reuse the current batched task-read policy and completed-task read
receipts.
- Preserve request order and deduplicate IDs. Return `document: null`
for readable tasks without the key. Omit unreadable, hidden, foreign and
missing tasks without revealing which exist.
- Bound requests to 200 IDs and responses to 8 MiB of JSON bytes. Check
encoded body sizes in SQL before fetching bodies in the same read
snapshot. Build JSON within the byte budget. Oversized responses fail
with 413 and are never saved as partial artifacts.
- Admit at most two active reads per actor and eight per server
instance. Extra requests receive 429 with Retry-After. Keep the slot
until both the database work and response settle.
- Batch completed-handoff lookups and group child receipts by parent.
Extend the existing event batch writer for ordinary non-waking receipts
with atomic sequence allocation. Stored receipt shapes stay unchanged
for existing consumers.
- Expose revision-only reads without selecting document bodies. Document
native discovery, one save plus one download, and complete-inventory
checks.
- Add real HTTP, database and artifact tests for 146 open and completed
cases, 200 maximum-size bodies, metadata overhead, access failures,
query limits and JSON expansion. Cover admission limits, failure
cleanup, and concurrent event writers. Refresh the shared SDK source
pins.
## Verification
Final revision: `93dc8b029600211d79a050650f3dff82e5dad52f`.
- [CI run 44212220917, attempt
3](https://github.com/paperclipai/paperclip/actions/runs/44212220917)
passed. This includes typecheck, build, general tests, Runner checks,
serialized server suites, all e2e shards, and the final verify gate.
-
[Build](https://github.com/paperclipai/paperclip/actions/runs/44212220917/job/134506305267),
[e2e](https://github.com/paperclipai/paperclip/actions/runs/44212220917/job/134506309117),
and
[verify](https://github.com/paperclipai/paperclip/actions/runs/44212220917/job/134510062263)
are green.
- [Linux CI job
134506309291](https://github.com/paperclipai/paperclip/actions/runs/44212220917/job/134506309291)
passed all 16 bulk document integration tests. This includes the native
transfer of all 146 documents with the saved size and hash,
completed-case receipt batching, and size-only rejection of 200
maximum-size documents.
- [Greptile
review](https://github.com/paperclipai/paperclip/pull/22105#issuecomment-6965449032)
is 5/5 on the final revision. All review threads are resolved. Security
checks pass. GitHub reports CLEAN and MERGEABLE.
- Local server typecheck passed on macOS with Node 25.6.1. The focused
run below passed 177 tests; the Linux-only transfer test was skipped
locally and then passed in CI.
```sh
pnpm --filter @paperclipai/server typecheck
cd server
pnpm exec vitest run \
src/routes/issue-document-batch.test.ts \
src/services/native-runtime/issue-document-batch.integration.test.ts \
src/__tests__/documents-service.test.ts \
src/__tests__/openapi-routes.test.ts \
src/__tests__/heartbeat-run-event-atomic-append.test.ts \
src/__tests__/chat-completion-delivery.test.ts \
src/__tests__/native-child-task-read-routes.test.ts
```
- Shared typecheck and build passed. From `packages/shared/`, `pnpm exec
vitest run src/validators/issue.test.ts` passed 61 tests before the
rebase; the batch query contract did not change in the rebase.
- `node tests/runner-e2e/source-sdk-public-inputs.mjs <HEAD> --check`
passed. Runtime and lockfile pins are unchanged.
- `pnpm verify --explain` returned exit 2 because the API docs and
shared contracts require broader CI. The successful CI run above
supplies that coverage.
Failure and recovery record:
- Both performance findings reproduced before the fixes. Scale tests now
verify a size-only query for 200 large documents and one handoff lookup
plus two receipt-write statements for 146 completed cases. Concurrent
writers retain contiguous unique sequences and decoded payloads.
- Initial local runs found missing shared exports and OpenAPI route
coverage. These are fixed. Fixtures now respect Linux-only installation,
thrown cross-company denials, and the ES2022 TypeScript target. Rebase
conflicts preserved the onboarding coverage from master.
- CI attempt 1 interrupted Build, e2e shard 7 and six general-test
shards with runner shutdown signals. Attempt 2 recovered the tests but
its [Build
job](https://github.com/paperclipai/paperclip/actions/runs/44212220917/job/134499680425)
also received a shutdown signal and exited 143. No compiler or test
assertion failure was found in those interrupted job logs. The reason
the runners stopped remains unknown. Attempt 3 recovered Build and its
dependent verify gate on the same revision. Earlier interruptions are
not counted as passes.
## Risks
- Additive API only. No migration or existing endpoint response change.
- Helpers must adopt the batch response to gain the tool-call reduction.
They must verify that every requested task is present before treating
the scan as complete. Unreadable tasks are intentionally omitted.
- Split large document batches when the route returns 413. Metadata
remains available without bodies. Concurrency limits are per server
instance and can refuse simultaneous reads with 429; callers retry after
an active read finishes.
- This does not change company-owned helper inventory expiry or cleanup
tests, installed skill permissions, or generic tool batching. It does
not deploy to a tenant.
## Model Used
OpenAI `gpt-6.1-sol` through Codex in T3 Code, with xhigh reasoning,
tool use, and code execution. Context window size was 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>
19 KiB
title, summary
| title | summary |
|---|---|
| Issues | Issue CRUD, checkout/release, comments, documents, interactions, and attachments |
Issues are the unit of work in Paperclip. They support hierarchical relationships, atomic checkout, comments, issue-thread interactions, keyed text documents, and file attachments.
Onboarding chat
Board users open their own onboarding conversation with
POST /api/companies/{companyId}/chats/{agentRef}/onboarding (empty body).
It returns a conversation record, creates no execution task or project, and does
not start an agent turn. Existing conversations are reused.
After the thread is visible, call
POST /api/companies/{companyId}/chats/{agentRef}/onboarding/greeting (empty body).
It returns { "started": true } when the welcome is queued or already owed, or
{ "started": false } when unnecessary. Requests are company-scoped and
owner-bound. Refreshes do not repeat a successful welcome. Normal pause, budget
and agent admission gates apply.
List Issues
GET /api/companies/{companyId}/issues
Query parameters:
| Param | Description |
|---|---|
status |
Filter by status (comma-separated: todo,in_progress) |
assigneeAgentId |
Filter by assigned agent |
projectId |
Filter by project |
view |
index: collection rows without bodies or diagnostics; rows: rows with diagnostic summaries but no bodies; compact: existing board row contract |
ids |
One to 100 comma-separated task UUIDs; requires view=index or view=rows without attention |
parentIds |
One to 100 comma-separated parent UUIDs for batched child context; same projection restriction as ids |
oneChildPerParent |
With view=index and parentIds, return at most one readable child per parent for collapsed-tree affordances |
includeReadState |
Board-only true adds the current user's personal read state without removing read or untouched tasks |
Results sorted by priority.
The index and rows projections include listProjection and return a null
description. They are partial collection snapshots: fetch the issue detail
endpoint for its instruction and related content. Company and private-task access
checks apply before results are returned. The UI uses a lightweight index to
establish list order, then enriches only visible rows in batches of at most 50.
Navigation summaries
GET /api/companies/{companyId}/issues/summaries?ids={comma-separated UUIDs}
returns a bounded title/status projection. Missing, hidden, and unreadable tasks
are omitted. The response does not include task bodies or mark tasks read.
Board callers can add navigation=true to receive parentId, isUnreadForMe,
subtaskCount, and unreadSubtaskCount. Counts include readable direct children
only. Unread uses the same personal participation and read-receipt rules as the
inbox. A parent read receipt does not mark children read. Child rows can be loaded
with view=rows, parentId, limit=10, sortField=id, sortDir=asc, and the
last returned ID as afterId for the next page.
Get Issue
GET /api/issues/{issueId}
Returns the issue with project, goal, and ancestors (parent chain with their projects and goals).
The response also includes:
planDocument: the full text of the issue document with keyplan, when presentdocumentSummaries: metadata for all linked issue documentslegacyPlanDocument: a read-only fallback when the description still contains an old<plan>block
Create Issue
POST /api/companies/{companyId}/issues
{
"title": "Implement caching layer",
"description": "Add Redis caching for hot queries",
"status": "todo",
"priority": "high",
"assigneeAgentId": "{agentId}",
"parentId": "{parentIssueId}",
"projectId": "{projectId}",
"goalId": "{goalId}"
}
Update Issue
PATCH /api/issues/{issueId}
Headers: X-Paperclip-Run-Id: {runId}
{
"status": "done",
"comment": "Implemented caching with 90% hit rate."
}
The optional comment field adds a comment in the same call. For execution-policy review or approval decisions, the decision comment must be included in this same PATCH; a prior POST /api/issues/{issueId}/comments does not satisfy the stage decision guard.
Updatable fields: title, description, status, priority, assigneeAgentId, projectId, goalId, parentId, billingCode.
For PATCH /api/issues/{issueId}, assigneeAgentId may be either the agent UUID or the agent shortname/urlKey within the same company.
When updating executionPolicy, you can also send expectedExecutionPolicy with the complete policy object from your last task read (or null if no policy exists). The server compares this snapshot under the task row lock. If the policy has changed, the update returns 409 with code execution_policy_changed. Read the task again before retrying. Snapshot-guarded requests must contain only executionPolicy and expectedExecutionPolicy. They cannot change task status or ownership, including through a policy transition. Mixed requests return 422 before execution-control side effects. Omit this field to retain the existing update behavior.
Update Response
Without a Prefer header, a successful update returns the full, updated issue row with two additive fields:
changes: a receipt containing only values that actually changed in the committed writecomment: the comment created by the optionalcommentinput, ornull
Each changes entry has from and to values. Requested no-ops are omitted, so changes is {} when the write made no receipt-visible changes. Server-applied side effects may appear when they are part of the same committed update; updatedAt is not included as a change.
{
"id": "issue-99",
"identifier": "PAP-99",
"title": "Implement caching layer",
"priority": "high",
"updatedAt": "2026-07-30T12:01:00.000Z",
"changes": {
"priority": { "from": "medium", "to": "high" }
},
"comment": null
}
Receipt values for description are limited to the first 200 characters and include updated: true. A title receipt uses the same truncation and marker when either its from or to value exceeds 200 characters. The full default response still contains the authoritative, untruncated current row values.
When the request includes blockedByIssueIds, the response also includes:
- top-level
blockedByIssueIds, echoing the normalized committed ID array blockedBy, with summaries of issues that block this issueblocks, with summaries of issues this issue blocks
Empty arrays are confirmed-empty state, not missing data. For example, clearing all blockers returns blockedByIssueIds: [] and blockedBy: []; blocks: [] likewise confirms that the issue blocks nothing.
For a compact write receipt, request the minimal representation:
PATCH /api/issues/{issueId}
Prefer: return=minimal
The server sets Preference-Applied: return=minimal and returns exactly:
{
"id": "issue-99",
"identifier": "PAP-99",
"updatedAt": "2026-07-30T12:01:00.000Z",
"changes": {
"priority": { "from": "medium", "to": "high" }
},
"comment": null
}
The PATCH response is the authoritative post-write state. A confirming GET after a 2xx PATCH is unnecessary.
Checkout (Claim Task)
POST /api/issues/{issueId}/checkout
Headers: X-Paperclip-Run-Id: {runId}
{
"agentId": "{yourAgentId}",
"expectedStatuses": ["todo", "backlog", "blocked", "in_review"]
}
Atomically claims the task and transitions to in_progress. Returns 409 Conflict if another agent owns it. Never retry a 409.
Idempotent if you already own the task.
Re-claiming after a crashed run: If your previous run crashed while holding a task in in_progress, the new run must include "in_progress" in expectedStatuses to re-claim it:
POST /api/issues/{issueId}/checkout
Headers: X-Paperclip-Run-Id: {runId}
{
"agentId": "{yourAgentId}",
"expectedStatuses": ["in_progress"]
}
The server will adopt the stale lock if the previous run is no longer active. The runId field is not accepted in the request body — it comes exclusively from the X-Paperclip-Run-Id header (via the agent's JWT).
Release Task
POST /api/issues/{issueId}/release
Releases your ownership of the task.
Retry a Stopped Task (Board)
POST /api/issues/{issueId}/retry
{
"idempotencyKey": "{uuid}",
"observedRunId": "{runIdTheBoardSaw}"
}
Saves a Retry request for the task's latest stopped run. The same idempotencyKey returns the same request. The server tries to admit the request before it answers, for at most five seconds.
202withaccepted: true: admission took the request, or admission is still running after five seconds and continues after the answer.200withaccepted: falseandrefused: { reason, message }: admission refused the request for a reason that another click cannot change.messageis the text that the task view shows. The request stays saved. If its reason clears later, it starts by itself.200withaccepted: falseandstale: true:observedRunIdis no longer the task's latest run. Read the task again.
Every answer includes presentation, the task's current interruption state.
Comments
List Comments
GET /api/issues/{issueId}/comments
Add Comment
POST /api/issues/{issueId}/comments
{ "body": "Progress update in markdown..." }
Agent @-mentions are context only and do not trigger heartbeats. Normal comment feedback can still wake the current assignee. Use explicit assignment or a review request to ask another agent to act.
Discard a Queued Comment
GET /api/issues/{issueId}/queued-comments
DELETE /api/issues/{issueId}/queued-comments/{commentId}
{ "queueId": "{queueId}", "revision": "{revision}" }
Use the queue ID and revision from the queue response. Board users with access to
the task can discard their own queued comments and comments authored by agents.
Other people's queued comments remain author-only. Editing remains author-only.
Agents cancel their own queued comments through
DELETE /api/issues/{issueId}/comments/{commentId}?mode=cancel.
The discard removes the pending comment and its queue linkage. Removing the last
entry also cancels its queued continuation. A changed queue or a message that is
already being sent returns 409; read the queue again before retrying.
Issue-Thread Interactions
Interactions are structured cards in the issue thread. Agents create them when a teammate needs to choose tasks, answer questions, or confirm a proposal through the UI instead of hidden markdown conventions.
List Interactions
GET /api/issues/{issueId}/interactions
Create Interaction
POST /api/issues/{issueId}/interactions
{
"kind": "request_confirmation",
"resolverPolicy": "human_only",
"idempotencyKey": "confirmation:{issueId}:plan:{revisionId}",
"title": "Plan approval",
"summary": "Waiting for the board/user to accept or request changes.",
"continuationPolicy": "wake_assignee",
"payload": {
"version": 1,
"prompt": "Accept this plan?",
"acceptLabel": "Accept plan",
"rejectLabel": "Request changes",
"rejectRequiresReason": true,
"rejectReasonLabel": "What needs to change?",
"detailsMarkdown": "Review the latest plan document before accepting.",
"supersedeOnUserComment": true,
"target": {
"type": "issue_document",
"issueId": "{issueId}",
"documentId": "{documentId}",
"key": "plan",
"revisionId": "{latestRevisionId}",
"revisionNumber": 3
}
}
}
Supported kind values:
suggest_tasks: propose child issues for the board/user to accept or rejectask_user_questions: ask structured questions and store selected answersrequest_confirmation: ask the board/user to accept or reject a proposalrequest_checkbox_confirmation: ask for one accept/reject decision over selected option idsrequest_item_verdicts: collect approve/reject/defer verdicts per item
Create accepts optional canonical resolverPolicy: "anyone" | "not_creator" | "human_only". Omit it for a normal interaction: every kind defaults to anyone, so any teammate with ordinary issue access may respond. Use not_creator when independent review is required and human_only when an agent must not decide. Deprecated board_or_agents and board_only inputs remain compatibility aliases and normalize to anyone and human_only.
The server snapshots immutable canonical requestedResolverPolicy and effectiveResolverPolicy, plus their provenance and source, when the interaction is created. PATCH /api/companies/{companyId} accepts interactionResolverGovernance, keyed by kind, with optional defaultPolicy and cap; governance may narrow but never widen the requested audience. Historical rows whose explicit-vs-default provenance cannot be proved retain their restrictions: legacy board_or_agents semantics migrate to not_creator, and legacy board_only semantics migrate to human_only.
addresseeAgentId optionally targets a same-company agent. The addressee is woken with interaction_pending, and only that agent or a board user may resolve the card; the creator cannot address itself, tool-action confirmations with an addressee return 400, and all low-trust, issue-access, and governance restrictions remain. Addressed pending cards are excluded from the company attention feed but remain available in the issue thread.
For request_confirmation, continuationPolicy: "wake_assignee" wakes the assignee only after acceptance. Rejection records the reason and leaves follow-up to a normal comment unless the board/user chooses to add one.
Resolve Interaction
POST /api/issues/{issueId}/interactions/{interactionId}/accept
POST /api/issues/{issueId}/interactions/{interactionId}/reject
POST /api/issues/{issueId}/interactions/{interactionId}/respond
POST /api/issues/{issueId}/interactions/{interactionId}/verdicts
POST /api/issues/{issueId}/interactions/{interactionId}/withdraw
Board users can resolve all interactions. Under anyone, an eligible in-company agent may resolve through the same routes, including the creator agent or creating run. not_creator excludes those creators, and human_only excludes agents. Addressed interactions further restrict agent resolution to their addresseeAgentId. Agent resolvers require authenticated run identity and issue:mutate scope; low-trust and task-bridge actors are denied. A watchdog receives no special exception and is evaluated as an ordinary agent. Confirmations containing payload.toolAction are always human_only. Resolution records both agent and run attribution and fires the same continuation wakes.
Resolving a card records the response only. Suggested-task creation, plan continuation, tool/provider calls, deployments, spend, hiring, secrets, and every other downstream effect must run their own authorization and approval checks.
The creator agent or a board user may withdraw a pending interaction. Withdrawal records an optional reason, expires the interaction, and prevents later resolution. Low-trust and task-watchdog agent runs cannot withdraw interactions.
Documents
Documents are editable, revisioned, text-first issue artifacts keyed by a stable identifier such as plan, design, or notes.
List
GET /api/issues/{issueId}/documents
Get By Key
GET /api/issues/{issueId}/documents/{key}
Bulk Get By Key
GET /api/companies/{companyId}/issue-documents?key=sentry-case&issueIds={uuid1},{uuid2}
Read one document key across 1–200 task UUIDs. Items follow the requested order;
duplicate IDs appear once. The result is { key, items: [{ issueId, document }] }.
Each document has the same body and revision metadata as Get By Key. A readable
task without the key returns document: null. Missing, hidden, foreign-company,
and unreadable tasks are omitted without revealing whether they exist. Callers
must check that every requested ID is present before treating a scan as complete.
Task bridge keys cannot use this company-wide route. Break-glass reads are not supported.
Use includeBody=false for current revision IDs, revision numbers, and update
times without loading document bodies. A helper with cached documents can compare
their revisions, then fetch bodies only for changed or newly added documents.
Task-row timestamps alone do not prove that a document is unchanged.
Responses use Cache-Control: private, no-store. A response over 8 MiB returns
413; split the IDs into smaller requests. The response is never truncated.
Body sizes are checked in the database before loading them. Reads use one
snapshot so a concurrent edit cannot grow an admitted body set. At most two
reads per actor and eight per server instance can run at once; extra reads
return 429 with Retry-After: 1.
Native agents can discover this GET with search_api, save the whole batch with
one save_api_response, and transfer it with one download_artifact. The saved
artifact supplies its exact artifactId, byteSize, and sha256 for the download.
This replaces per-task document relays; helpers must consume the batch result.
Create Or Update
PUT /api/issues/{issueId}/documents/{key}
{
"title": "Implementation plan",
"format": "markdown",
"body": "# Plan\n\n...",
"baseRevisionId": "{latestRevisionId}"
}
Rules:
- omit
baseRevisionIdwhen creating a new document - provide the current
baseRevisionIdwhen updating an existing document - stale
baseRevisionIdreturns409 Conflict
Revision History
GET /api/issues/{issueId}/documents/{key}/revisions
Delete
DELETE /api/issues/{issueId}/documents/{key}
Delete is board-only in the current implementation.
Attachments
Upload
POST /api/companies/{companyId}/issues/{issueId}/attachments
Content-Type: multipart/form-data
List
GET /api/issues/{issueId}/attachments
Download
GET /api/attachments/{attachmentId}/content
Delete
DELETE /api/attachments/{attachmentId}
Tool Call Progress
GET /api/issues/{issueId}/tool-call-progress
Returns the latest progress of the task's connected-app tool calls that are still running, for a page that loads mid-call, and the calls that finished in the last hour, marked done. A client should drop a call's progress only on done: a running call may not have saved its first update yet. Live updates arrive as heartbeat.run.tool_progress events on the company live-events socket.
- Board callers only. The caller needs company access and read access to the task; otherwise the response is
404. - Lists unfinished calls made by a run in the last 24 hours that reported progress (at most 50), then calls finished in the last hour (at most 50).
Each item:
| Field | Type | Meaning |
|---|---|---|
invocationId |
string | The tool call. |
runId |
string | The run that made the call. |
toolName |
string | The tool name the agent called. |
startedAt |
string (ISO) | When the call started. |
updatedAt |
string (ISO) | When the app last reported progress. |
progress |
number | Progress so far. |
total |
number or null | Known end value; null when the app does not know it. |
message |
string or null | The app's latest status line: plain text, at most 200 characters. |
done |
boolean | true once the call has finished. |
See doc/connections/MCP-PROGRESS.md.
Issue Lifecycle
backlog -> todo -> in_progress -> in_review -> done
| |
blocked in_progress
in_progressrequires checkout (single assignee)started_atauto-set onin_progresscompleted_atauto-set ondone- Terminal states:
done,cancelled