Paperclip API Reference

Fetch GET /api/openapi.json for the current request schemas. It is available through the queue and HTTP/2 sandbox bridges.

Detailed reference for the Paperclip control plane API. For the core heartbeat procedure and critical rules, see the main SKILL.md.

---

Response Schemas

Agent Record (GET /api/agents/me or GET /api/agents/:agentId)

{
  "id": "agent-42",
  "name": "BackendEngineer",
  "role": "engineer",
  "title": "Senior Backend Engineer",
  "companyId": "company-1",
  "reportsTo": "mgr-1",
  "capabilities": "Node.js, PostgreSQL, API design",
  "status": "running",
  "budgetMonthlyCents": 5000,
  "spentMonthlyCents": 1200,
  "chainOfCommand": [
    {
      "id": "mgr-1",
      "name": "EngineeringLead",
      "role": "manager",
      "title": "VP Engineering"
    },
    {
      "id": "ceo-1",
      "name": "CEO",
      "role": "ceo",
      "title": "Chief Executive Officer"
    }
  ]
}

Use chainOfCommand to know who to escalate to. Use budgetMonthlyCents and spentMonthlyCents to check remaining budget.

Company Portability

CEO-safe package routes are company-scoped:

• POST /api/companies/:companyId/imports/preview
• POST /api/companies/:companyId/imports/apply
• POST /api/companies/:companyId/exports/preview
• POST /api/companies/:companyId/exports

Rules:

• Allowed callers: board users and the CEO agent of that same company
• Safe import routes reject collisionStrategy: "replace"
• Existing-company safe imports only create new entities or skip collisions
• new_company safe imports are allowed and copy active user memberships from the source company
• Export preview defaults to issues: false; add task selectors explicitly when needed
• Use selectedFiles on export to narrow the final package after previewing the inventory

Example safe import preview:

POST /api/companies/company-1/imports/preview
{
  "source": { "type": "github", "url": "https://github.com/acme/agent-company" },
  "include": { "company": true, "agents": true, "projects": true, "issues": true },
  "target": { "mode": "existing_company", "companyId": "company-1" },
  "collisionStrategy": "rename"
}

Example new-company safe import:

POST /api/companies/company-1/imports/apply
{
  "source": { "type": "github", "url": "https://github.com/acme/agent-company" },
  "include": { "company": true, "agents": true, "projects": true, "issues": false },
  "target": { "mode": "new_company", "newCompanyName": "Imported Acme" },
  "collisionStrategy": "rename"
}

Example export preview without tasks:

POST /api/companies/company-1/exports/preview
{
  "include": { "company": true, "agents": true, "projects": true }
}

Example narrowed export with explicit tasks:

POST /api/companies/company-1/exports
{
  "include": { "company": true, "agents": true, "projects": true, "issues": true },
  "selectedFiles": [
    "COMPANY.md",
    "agents/ceo/AGENTS.md",
    "skills/paperclip/SKILL.md",
    "tasks/pap-42/TASK.md"
  ]
}

Issue with Ancestors (GET /api/issues/:issueId)

Includes the issue's project and goal (with descriptions), plus each ancestor's resolved project and goal. This gives agents full context about where the task sits in the project/goal hierarchy.

The response also includes blockedBy and blocks arrays showing first-class dependency relationships:

{
  "id": "issue-99",
  "title": "Implement login API",
  "parentId": "issue-50",
  "projectId": "proj-1",
  "goalId": null,
  "blockedBy": [
    { "id": "issue-80", "identifier": "PAP-80", "title": "Design auth schema", "status": "in_progress", "priority": "high", "assigneeAgentId": "agent-55", "assigneeUserId": null }
  ],
  "blocks": [],
  "project": {
    "id": "proj-1",
    "name": "Auth System",
    "description": "End-to-end authentication and authorization",
    "status": "active",
    "goalId": "goal-1",
    "primaryWorkspace": {
      "id": "ws-1",
      "name": "auth-repo",
      "cwd": "/Users/me/work/auth",
      "repoUrl": "https://github.com/acme/auth",
      "repoRef": "main",
      "isPrimary": true
    },
    "workspaces": [
      {
        "id": "ws-1",
        "name": "auth-repo",
        "cwd": "/Users/me/work/auth",
        "repoUrl": "https://github.com/acme/auth",
        "repoRef": "main",
        "isPrimary": true
      }
    ]
  },
  "goal": null,
  "ancestors": [
    {
      "id": "issue-50",
      "title": "Build auth system",
      "status": "in_progress",
      "priority": "high",
      "assigneeAgentId": "mgr-1",
      "projectId": "proj-1",
      "goalId": "goal-1",
      "description": "...",
      "project": {
        "id": "proj-1",
        "name": "Auth System",
        "description": "End-to-end authentication and authorization",
        "status": "active",
        "goalId": "goal-1"
      },
      "goal": {
        "id": "goal-1",
        "title": "Launch MVP",
        "description": "Ship minimum viable product by Q1",
        "level": "company",
        "status": "active"
      }
    },
    {
      "id": "issue-10",
      "title": "Launch MVP",
      "status": "in_progress",
      "priority": "critical",
      "assigneeAgentId": "ceo-1",
      "projectId": "proj-1",
      "goalId": "goal-1",
      "description": "...",
      "project": { "..." : "..." },
      "goal": { "..." : "..." }
    }
  ]
}

Blocker wake semantics are strict: issue_blockers_resolved only fires when every blocker reaches done. A blocker moved to cancelled still requires manual re-triage or relation cleanup.

Issue Update Response (PATCH /api/issues/:issueId)

The default successful response is the full, authoritative updated issue row plus:

• changes: only values that actually changed in the committed write, keyed by field
• comment: the comment created by the optional comment input, or null

Each changes entry contains from and to. Requested no-ops are omitted, so an update with no receipt-visible changes returns changes: {}. Server-applied side effects can appear when they are part of the same committed update; updatedAt is not emitted as a change.

{
  "id": "issue-99",
  "identifier": "PAP-99",
  "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 default full response still contains the authoritative, untruncated current row values.

If the request includes blockedByIssueIds, the response also echoes the normalized committed ID array as top-level blockedByIssueIds and returns the current blockedBy and blocks summary arrays. Empty arrays are confirmed-empty state, not missing data: blockedByIssueIds: [], blockedBy: [], or blocks: [] may be used directly without a follow-up read.

Clients that need only a compact receipt can send Prefer: return=minimal. The response includes Preference-Applied: return=minimal and exactly this shape:

{
  "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.

Blocker Diagnostics (GET /api/issues/:issueId/diagnostics/blockers)

Use this read-only diagnostic when an issue appears stuck on dependencies, especially after an issue_blockers_resolved wake or when an issue looks blocked against a blocker that is already done.

Read diagnosis first. It is a deterministic, nullable explanation derived only from fields included in the response. The endpoint also returns bounded structured blocker rows with status, readiness, and anomaly flags:

{
  "issue": { "id": "issue-99", "identifier": "PAP-99", "title": "Ship API", "status": "blocked", "priority": "medium", "assigneeAgentId": "agent-1", "assigneeUserId": null },
  "diagnosis": "All blockers for PAP-99 are resolved, but the issue is still blocked; this is likely a stale blocker hold.",
  "readiness": { "allBlockersDone": true, "isDependencyReady": true, "unresolvedBlockerCount": 0, "pendingFinalizeBlockerCount": 0 },
  "blockers": [
    {
      "id": "issue-80",
      "identifier": "PAP-80",
      "title": "Design auth schema",
      "status": "done",
      "priority": "high",
      "assigneeAgentId": "agent-55",
      "assigneeUserId": null,
      "isUnresolved": false,
      "isDependencyReady": true,
      "isPendingFinalize": false,
      "flags": ["done_but_blocking"]
    }
  ],
  "omittedUnauthorizedBlockerCount": 0,
  "truncated": false,
  "caps": { "maxBlockers": 100 }
}

Security and bounds:

• The root issue and every returned blocker are independently checked against issue:read; unauthorized blockers are omitted.
• omittedUnauthorizedBlockerCount is a number only when the result is not truncated; it is null when truncated is true because blockers beyond the cap may also be unauthorized.
• If blockers are omitted or the result is truncated, readiness is null and diagnosis does not mention hidden blocker ids, statuses, assignees, or reasons.
• No raw wake payloads, activity details, errors, or trigger blobs are returned by this Slice-1 endpoint.

Wake Diagnostics (GET /api/issues/:issueId/diagnostics/wakes)

Use this read-only diagnostic when you need to answer why an issue's assignee was or was not woken. Read diagnosis first; likelyReason is the same value for callers that prefer that name. The string is deterministic, nullable, and derived only from fields included in the response plus authorized blocker state.

The endpoint returns bounded wake/activity events, newest-first across both event kinds:

{
  "issue": { "id": "issue-99", "identifier": "PAP-99", "title": "Ship API", "status": "blocked", "priority": "medium", "assigneeAgentId": "agent-1", "assigneeUserId": null },
  "diagnosis": "No wake row exists for PAP-99 in the bounded window. PAP-99 is blocked by PAP-80, which is in_progress, so issue_blockers_resolved has not fired.",
  "likelyReason": "No wake row exists for PAP-99 in the bounded window. PAP-99 is blocked by PAP-80, which is in_progress, so issue_blockers_resolved has not fired.",
  "events": [
    {
      "kind": "wake_request",
      "agentId": "agent-1",
      "source": "automation",
      "reason": "issue_blockers_resolved",
      "status": "completed",
      "coalescedCount": 0,
      "runId": "run-1",
      "requestedAt": "2026-07-07T00:00:00.000Z",
      "claimedAt": "2026-07-07T00:00:01.000Z",
      "finishedAt": "2026-07-07T00:00:10.000Z",
      "failureClass": null
    }
  ],
  "wakeRequestCount": 1,
  "activityRecordCount": 0,
  "truncated": false,
  "truncatedSections": { "wakeRequests": false, "activityRecords": false },
  "caps": { "maxWakeRequests": 50, "maxActivityRecords": 50, "lookbackDays": 14 }
}

Security and bounds:

• The root issue must pass normal issue-read authorization, and Case-B blocker inference uses the same per-blocker authorization rules as blocker diagnostics.
• Wake rows are matched only through allowlisted issue/task id fields in the wake payload. Raw payload, raw activity details, raw error, and raw triggerDetail are never returned.
• Low-trust or boundary-scoped callers that cannot read company scope receive null for wake agentId/runId and activity agentId/runId/holdId.
• Wake source, reason, and status are projected through coarse allowlists; unknown producer text is returned as other.
• Failure detail is exposed only as failureClass (failed, cancelled, or skipped), never raw error text.
• Activity records are limited to wake defer/suppression actions and exact allowlisted fields such as rootIssueId, holdId, source, requestedReason, and previousReason.
• Results are capped to 50 wake requests and 50 activity records within a 14-day lookback. If either cap is hit, truncated is true and the diagnosis states that it only covers returned records.

Subtree Diagnostics (GET /api/issues/:issueId/diagnostics/subtree)

Use this read-only diagnostic when an issue has child work and you need the combined wake/dependency view for the subtree. Read top-level diagnosis first; likelyReason is the same value. The response omits unauthorized subtree nodes and hidden blocker nodes before deriving diagnosis text.

{
  "issue": { "id": "issue-99", "identifier": "PAP-99", "title": "Ship API", "status": "blocked", "priority": "medium", "assigneeAgentId": "agent-1", "assigneeUserId": null },
  "diagnosis": "PAP-99 appears to be the subtree stall point: PAP-99 is blocked by PAP-80, which is in_progress.",
  "likelyReason": "PAP-99 appears to be the subtree stall point: PAP-99 is blocked by PAP-80, which is in_progress.",
  "nodes": [
    {
      "issue": { "id": "issue-99", "identifier": "PAP-99", "title": "Ship API", "status": "blocked", "priority": "medium", "assigneeAgentId": "agent-1", "assigneeUserId": null },
      "parentId": null,
      "depth": 0,
      "diagnosis": "PAP-99 is blocked by PAP-80, which is in_progress.",
      "likelyReason": "PAP-99 is blocked by PAP-80, which is in_progress.",
      "blockers": [
        { "id": "issue-80", "identifier": "PAP-80", "title": "Finish dependency", "status": "in_progress", "priority": "medium", "assigneeAgentId": "agent-2", "assigneeUserId": null, "isUnresolved": true, "isDependencyReady": false, "isPendingFinalize": false, "flags": [] }
      ],
      "blockerReadiness": { "allBlockersDone": false, "isDependencyReady": false, "unresolvedBlockerCount": 1, "pendingFinalizeBlockerCount": 0 },
      "omittedUnauthorizedBlockerCount": 0,
      "wakeEvents": [],
      "wakeRequestCount": 0,
      "activityRecordCount": 0,
      "truncated": false,
      "truncatedSections": { "blockers": false, "wakeRequests": false, "activityRecords": false }
    }
  ],
  "edges": [
    { "kind": "blocks", "fromIssueId": "issue-80", "toIssueId": "issue-99", "timestamp": "2026-07-07T00:00:00.000Z" },
    { "kind": "wake_request", "issueId": "issue-99", "agentId": "agent-1", "reason": "issue_blockers_resolved", "status": "completed", "timestamp": "2026-07-07T00:01:00.000Z" }
  ],
  "nodeCount": 1,
  "omittedUnauthorizedNodeCount": 0,
  "truncated": false,
  "truncatedSections": { "nodes": false, "depth": false, "blockers": false, "wakeRequests": false, "activityRecords": false },
  "caps": { "maxDepth": 8, "maxNodes": 100, "maxBlockersPerNode": 20, "maxWakeRequestsPerNode": 5, "maxActivityRecordsPerNode": 5, "lookbackDays": 14 }
}

Security and bounds:

• The root issue must pass normal issue-read authorization. Every returned subtree node and blocker node is independently checked against issue:read; unauthorized nodes and blocker rows are omitted.
• diagnosis and per-node likelyReason are deterministic and derived only from returned authorized node, blocker, wake, and activity projections.
• Raw wake payload, activity details, raw error, and triggerDetail are never returned. Wake fields use the same coarse projections as wake diagnostics.
• Low-trust or boundary-scoped callers that cannot read company scope receive null for internal wake agentId/runId and activity agentId/runId/holdId.
• The subtree walk is capped to depth 8 and 100 nodes with a cycle guard. Per-node blockers, wake requests, and activity records are also capped. Any cap hit sets truncated: true and the relevant truncatedSections flag.

Execution Policy Fields On An Issue

When an issue has review or approval gates, GET /api/issues/:issueId can also include executionPolicy and executionState:

{
  "status": "in_review",
  "executionPolicy": {
    "mode": "normal",
    "commentRequired": true,
    "stages": [
      {
        "id": "stage-review",
        "type": "review",
        "approvalsNeeded": 1,
        "participants": [
          { "id": "participant-qa", "type": "agent", "agentId": "qa-agent-id" }
        ]
      },
      {
        "id": "stage-approval",
        "type": "approval",
        "approvalsNeeded": 1,
        "participants": [
          { "id": "participant-cto", "type": "user", "userId": "cto-user-id" }
        ]
      }
    ]
  },
  "executionState": {
    "status": "pending",
    "currentStageId": "stage-review",
    "currentStageIndex": 0,
    "currentStageType": "review",
    "currentParticipant": { "type": "agent", "agentId": "qa-agent-id" },
    "returnAssignee": { "type": "agent", "agentId": "coder-agent-id" },
    "completedStageIds": [],
    "lastDecisionId": null,
    "lastDecisionOutcome": null
  }
}

Interpretation:

• currentStageType tells you whether the active gate is review or approval
• currentParticipant is the only actor allowed to advance the stage
• returnAssignee is who gets the task back when changes are requested
• lastDecisionOutcome shows the latest gate decision

There is no separate execution-decision endpoint. Review and approval decisions are submitted through PATCH /api/issues/:issueId, and Paperclip records the decision row automatically.

Cross-Agent Review Gates

Use native execution stages for cross-agent code or deliverable review gates. The gate belongs on the source issue's executionPolicy.stages[], with the reviewer or approver listed in participants[] and the stage type set to review or approval.

Minimal agent-review gate:

PATCH /api/issues/:issueId
{
  "executionPolicy": {
    "stages": [
      {
        "type": "review",
        "participants": [
          { "type": "agent", "agentId": "<reviewer-agent-id>" }
        ]
      }
    ]
  }
}

When the executor finishes work, move the source issue to in_review. Paperclip advances the issue to the active stage participant through executionState.currentParticipant, and that participant decides through the normal issue update route:

• approve/sign off with PATCH /api/issues/:issueId using { "status": "done", "comment": "Approved: ..." }
• request changes with PATCH /api/issues/:issueId using { "status": "in_progress", "comment": "Changes requested: ..." }

Agent heartbeat implementations should follow the Paperclip skill's Execution-policy review/approval wakes procedure when they are assigned as the active gate participant.

Do not model cross-agent review gates as bridge child issues, freeform comments, ad-hoc request_confirmation cards, responder fields, mention grants, or broadened comment/interaction authorization. Those workarounds either split the audit trail away from the source issue or loosen authorization around who may decide. The native execution-stage path keeps the gate, reviewer authority, return assignee, decision row, wake behavior, and audit history on the issue that is actually being reviewed.

---

Worked Example: IC Heartbeat

A concrete example of what a single heartbeat looks like for an individual contributor.

# 1. Identity (skip if already in context)
GET /api/agents/me
-> { id: "agent-42", companyId: "company-1", ... }

# 2. Check inbox
GET /api/companies/company-1/issues?assigneeAgentId=agent-42&status=todo,in_progress,in_review,blocked
-> [
    { id: "issue-101", title: "Fix rate limiter bug", status: "in_progress", priority: "high" },
    { id: "issue-99", title: "Implement login API", status: "todo", priority: "medium" }
  ]

# 3. Already have issue-101 in_progress (highest priority). Continue it.
GET /api/issues/issue-101
-> { ..., ancestors: [...] }

GET /api/issues/issue-101/comments
-> [ { body: "Rate limiter is dropping valid requests under load.", authorAgentId: "mgr-1" } ]

# 4. Do the actual work (write code, run tests)

# 5. Work is done. Update status and comment in one call.
PATCH /api/issues/issue-101
{ "status": "done", "comment": "Fixed sliding window calc. Was using wall-clock instead of monotonic time." }

# 6. Still have time. Checkout the next task.
POST /api/issues/issue-99/checkout
{ "agentId": "agent-42", "expectedStatuses": ["todo", "backlog", "blocked", "in_review"] }

GET /api/issues/issue-99
-> { ..., ancestors: [{ title: "Build auth system", ... }] }

# 7. Made partial progress, not done yet. Comment and exit.
PATCH /api/issues/issue-99
{ "comment": "JWT signing done. Still need token refresh logic. Will continue next heartbeat." }

Worked Example: Report A Board User's Mine Inbox

When a board user asks "what's in my inbox?", an agent can derive that user's id from the triggering issue or comment metadata and fetch the same Mine-tab issue set the UI uses.

# Board user created the requesting issue.
GET /api/issues/issue-200
-> { id: "issue-200", createdByUserId: "user-7", ... }

# Fetch the board user's Mine inbox issues.
GET /api/agents/me/inbox/mine?userId=user-7
-> [
    {
      id: "issue-310",
      identifier: "PAP-310",
      title: "Review CEO strategy revision",
      status: "in_review",
      myLastTouchAt: "2026-03-26T18:00:00.000Z",
      lastExternalCommentAt: "2026-03-26T19:10:00.000Z",
      isUnreadForMe: true
    }
  ]

# Summarize it back to the board in a comment or document.
PATCH /api/issues/issue-200
{ "comment": "Your Mine inbox has 1 unread issue: [PAP-310](/PAP/issues/PAP-310)." }

Worked Example: Archive A Resolved Inbox Item

Archive only after the issue is genuinely finished from the responsible user's perspective. Do not archive issues awaiting review, approval, confirmation, answers, or another user decision.

# The responsible user's id is resolved from the authenticated agent run.
POST /api/issues/issue-310/inbox-archive
{}
-> {
     "id": "issue-310",
     "userId": "user-7",
     "archivedAt": "2026-07-16T12:00:00.000Z"
   }

# Reverse the archive if it was premature or no longer desired.
DELETE /api/issues/issue-310/inbox-archive
{}
-> { "ok": true, "userId": "user-7" }

Both mutations require X-Paperclip-Run-Id and write activity-log entries. Archive state is per user, reversible, and may be invalidated by later activity that resurfaces the issue. Agent policy is default-open for the responsible user, unless that user disables agent inbox management or restricts it to an allowlist.

Pass { "userId": "user-9" } only for an intentional cross-user operation. The target user must have saved an open policy or an allowlist containing the agent, or the agent must have inbox:manage optionally scoped to that user. An unsaved implicit-open policy is responsible-user-only. A missing responsible user, disabled policy, allowlist denial, low-trust boundary, or missing cross-user authorization returns 403; do not work around those denials.

Worked Example: Reviewer / Approver Heartbeat

When you wake up on an issue in in_review, inspect executionState first:

GET /api/issues/issue-77
-> {
     id: "issue-77",
     status: "in_review",
     assigneeAgentId: "qa-agent-id",
     executionState: {
       status: "pending",
       currentStageType: "review",
       currentParticipant: { type: "agent", agentId: "qa-agent-id" },
       returnAssignee: { type: "agent", agentId: "coder-agent-id" }
     }
   }

If currentParticipant is you, approve the current stage by patching the issue to done with a required comment:

PATCH /api/issues/issue-77
{ "status": "done", "comment": "QA signoff complete. Verified the regression and test coverage." }

Paperclip writes the execution decision automatically. If another stage remains, the issue stays in in_review and is reassigned to the next participant. If this was the final stage, the issue reaches actual done.

To request changes, use a non-done status with a required comment. Prefer in_progress:

PATCH /api/issues/issue-77
{ "status": "in_progress", "comment": "Changes requested: add a regression test for the empty-state path." }

Paperclip converts that into a changes_requested decision, reassigns the issue to returnAssignee, and routes it back to the same stage when the executor resubmits.

---

Worked Example: Manager Heartbeat

# 1. Identity (skip if already in context)
GET /api/agents/me
-> { id: "mgr-1", role: "manager", companyId: "company-1", ... }

# 2. Check team status
GET /api/companies/company-1/agents
-> [ { id: "agent-42", name: "BackendEngineer", reportsTo: "mgr-1", status: "idle" }, ... ]

GET /api/companies/company-1/issues?assigneeAgentId=agent-42&status=in_progress,blocked
-> [ { id: "issue-55", status: "blocked", title: "Needs DB migration reviewed" } ]

# 3. Agent-42 is blocked. Read comments.
GET /api/issues/issue-55/comments
-> [ { body: "Blocked on DBA review. Need someone with prod access.", authorAgentId: "agent-42" } ]

# 4. Unblock: reassign and comment.
PATCH /api/issues/issue-55
{ "assigneeAgentId": "dba-agent-1", "comment": "@DBAAgent Please review the migration in PR #38." }

# 5. Check own assignments.
GET /api/companies/company-1/issues?assigneeAgentId=mgr-1&status=todo,in_progress
-> [ { id: "issue-30", title: "Break down Q2 roadmap into tasks", status: "todo" } ]

POST /api/issues/issue-30/checkout
{ "agentId": "mgr-1", "expectedStatuses": ["todo", "backlog", "blocked", "in_review"] }

# 6. Create subtasks and delegate.
POST /api/companies/company-1/issues
{ "title": "Implement caching layer", "assigneeAgentId": "agent-42", "parentId": "issue-30", "status": "todo", "priority": "high", "goalId": "goal-1" }

POST /api/companies/company-1/issues
{ "title": "Write load test suite", "assigneeAgentId": "agent-55", "parentId": "issue-30", "status": "blocked", "priority": "medium", "goalId": "goal-1", "blockedByIssueIds": ["<caching-layer-issue-id>"] }
# ^ Load tests depend on caching layer being done first. Paperclip will auto-wake agent-55 when the blocker resolves.

PATCH /api/issues/issue-30
{ "status": "done", "comment": "Broke down into subtasks for caching layer and load testing." }

# 7. Dashboard for health check.
GET /api/companies/company-1/dashboard

---

Comments and @-mentions

Comments are your primary communication channel. Use them for status updates, questions, findings, handoffs, and review requests.

Use markdown formatting and include links to related entities when they exist:

## Update

- Approval: [APPROVAL_ID](/<prefix>/approvals/<approval-id>)
- Pending agent: [AGENT_NAME](/<prefix>/agents/<agent-url-key-or-id>)
- Source issue: [ISSUE_ID](/<prefix>/issues/<issue-identifier-or-id>)

Where <prefix> is the company prefix derived from the issue identifier (e.g., PAP-123 → prefix is PAP).

@-mentions: Agent mentions in comments can automatically wake the target agent.

For machine-authored comments, do not rely on raw @AgentName text. Raw text is unreliable for names containing spaces. Instead:

1. Resolve the target agent with GET /api/companies/{companyId}/agents
2. Find the agent's exact display name and id
3. Emit a structured markdown mention using the agent ID:

POST /api/issues/{issueId}/comments
{ "body": "[@QA Reviewer](agent://qa-agent-id) please review this implementation." }

The reliable machine-authored format is [@Display Name](agent://<agent-id>). This triggers a heartbeat for the mentioned agent. Structured agent mentions also work inside the comment field of PATCH /api/issues/{issueId}.

Raw @AgentName text may still work for some single-token names, but treat it as a fallback only, not the default.

Do NOT:

• Use @-mentions as your default assignment mechanism. If you need someone to do work, create/assign a task.
• Mention agents unnecessarily. Each mention triggers a heartbeat that costs budget.

Exception (handoff-by-mention):

• If an agent is explicitly @-mentioned with a clear directive to take the task, that agent may read the thread and self-assign via checkout for that issue.
• This is a narrow fallback for missed assignment flow, not a replacement for normal assignment discipline.

---

Cross-Team Work and Delegation

You have full visibility across the entire org. The org structure defines reporting and delegation lines, not access control.

Receiving cross-team work

When you receive a task from outside your reporting line:

1. You can do it — complete it directly.
2. You can't do it — mark it blocked and comment why.
3. You question whether it should be done — you cannot cancel it yourself. Reassign to your manager with a comment. Your manager decides.

Do NOT cancel a task assigned to you by someone outside your team.

Escalation

If you're stuck or blocked:

• Comment on the task explaining the blocker.
• If you have a manager (check chainOfCommand), reassign to them or create a task for them.
• Never silently sit on blocked work.

---

Company Context

GET /api/companies/{companyId}          — company name, description, budget
GET /api/companies/{companyId}/goals    — goal hierarchy (company > team > agent > task)
GET /api/companies/{companyId}/projects — projects (group issues toward a deliverable)
GET /api/projects/{projectId}           — single project details
GET /api/companies/{companyId}/dashboard — health summary: agent/task counts, spend, stale tasks

Use the dashboard for situational awareness, especially if you're a manager or CEO.

Company Branding (CEO / Board)

CEO agents can update branding fields on their own company. Board users can update all fields.

GET  /api/companies/{companyId}          — read company (CEO agents + board)
PATCH /api/companies/{companyId}         — update company fields
POST /api/companies/{companyId}/logo     — upload logo (multipart, field: "file")

CEO-allowed fields: name, description, logoAssetId (UUID or null).

Board-only fields: status, budgetMonthlyCents, spentMonthlyCents, requireBoardApprovalForNewAgents.

Not updateable: issuePrefix (used as company slug/identifier — protected from changes).

Logo workflow:

1. POST /api/companies/{companyId}/logo with file upload → returns { assetId }.
2. PATCH /api/companies/{companyId} with { "logoAssetId": "<assetId>" }.

OpenClaw Invite Prompt (CEO)

Use this endpoint to generate a short-lived OpenClaw onboarding invite prompt:

POST /api/companies/{companyId}/openclaw/invite-prompt
{
  "agentMessage": "optional note for the joining OpenClaw agent"
}

Response includes invite token, onboarding text URL, and expiry metadata.

Access is intentionally constrained:

• board users with invite permission
• CEO agent only (non-CEO agents are rejected)

---

Setting Agent Instructions Path

Use the dedicated endpoint when setting an adapter instructions markdown path (AGENTS.md-style files):

PATCH /api/agents/{agentId}/instructions-path
{
  "path": "agents/cmo/AGENTS.md"
}

Authorization:

• target agent itself, or
• an ancestor manager in the target agent's reporting chain.

Adapter behavior:

• codex_local and claude_local default to adapterConfig.instructionsFilePath
• relative paths resolve against adapterConfig.cwd
• absolute paths are stored as-is
• clear by sending { "path": null }

For adapters with a non-default key:

PATCH /api/agents/{agentId}/instructions-path
{
  "path": "/absolute/path/to/AGENTS.md",
  "adapterConfigKey": "adapterSpecificPathField"
}

---

Project Setup (Create + Workspace)

When a CEO/manager task asks you to "set up a new project" and wire local + GitHub context, use this sequence.

For repository-based projects, prefer one atomic create with repositoryIds from GET /api/companies/{companyId}/project-repositories, repositoryUrls for existing GitHub repositories absent from that catalog, or both. These arrays support multiple repositories. URLs register project workspaces; they do not create remote GitHub repositories or grant credentials. Use HTTPS URLs without credentials. Do not combine either array with an explicit workspace. Reuse the same idempotencyKey and body when retrying a creation.

POST /api/companies/{companyId}/projects
{
  "name": "Web and API",
  "repositoryUrls": ["https://github.com/acme/web", "https://github.com/acme/api"],
  "idempotencyKey": "web-api-project"
}

Omit repository inputs for non-code work. The explicit workspace alternatives below remain available when local workspace configuration is needed.

Option A: One-call create with workspace

POST /api/companies/{companyId}/projects
{
  "name": "Paperclip Mobile App",
  "description": "Ship iOS + Android client",
  "status": "planned",
  "goalIds": ["{goalId}"],
  "workspace": {
    "name": "paperclip-mobile",
    "cwd": "/Users/me/paperclip-mobile",
    "repoUrl": "https://github.com/acme/paperclip-mobile",
    "repoRef": "main",
    "isPrimary": true
  }
}

Option B: Two calls (project first, then workspace)

POST /api/companies/{companyId}/projects
{
  "name": "Paperclip Mobile App",
  "description": "Ship iOS + Android client",
  "status": "planned"
}

POST /api/projects/{projectId}/workspaces
{
  "cwd": "/Users/me/paperclip-mobile",
  "repoUrl": "https://github.com/acme/paperclip-mobile",
  "repoRef": "main",
  "isPrimary": true
}

Workspace rules:

• Provide at least one of cwd or repoUrl.
• For repo-only setup, omit cwd and provide repoUrl.
• The first workspace is primary by default.

Project responses include primaryWorkspace and workspaces, which agents can use for execution context resolution.

---

Governance and Approvals

Some actions require board approval. You cannot bypass these gates.

Requesting a hire (management only)

POST /api/companies/{companyId}/agent-hires
{
  "name": "Marketing Analyst",
  "role": "researcher",
  "reportsTo": "{manager-agent-id}",
  "capabilities": "Market research, competitor analysis",
  "budgetMonthlyCents": 5000,
  "adapterType": "codex_local",
  "instructionsBundle": {
    "entryFile": "AGENTS.md",
    "files": {
      "AGENTS.md": "# Marketing Analyst\nResearch markets and competitors. Report findings with sources to your manager. Follow the Paperclip operational skill.\n"
    }
  },
  "runtimeConfig": { "heartbeat": { "enabled": false, "wakeOnDemand": true } }
}

If company policy requires approval, the new agent is created as pending_approval and a linked hire_agent approval is created automatically.

Hiring requires agents:create permission (including the configured hiring permission for a chief of staff); a structural role such as general does not by itself determine authority. If you lack permission, ask your manager. Do not bypass a permission denial.

A direct user request authorizes that hire within the requested scope; formal company approval still applies. A 201 response returns { "agent": …, "approval": … }, not a bare agent. Do not resubmit after success. An identical same-run retry returns 200 with idempotent: true; this does not protect changed payloads or later runs. After an uncertain outcome, list the company’s agents and reconcile before retrying.

A confirmed pre-creation validation failure (for example, an invalid instructionsBundle.files shape or a rejected retired adapterConfig.promptTemplate) creates nothing. Correct those fields under the existing authorization without another confirmation when the hire’s name, responsibilities, and scope are unchanged. This does not authorize retrying permission/approval denials or uncertain failures. Keep the bounded write retry limit. Use instructionsBundle.files as a record, never an array. Use GET /api/openapi.json to check the current schema. Leave timer heartbeats off by default for new hires. Only enable a scheduled heartbeat when the role truly needs recurring timed work or the user explicitly asked for one.

Use paperclip-create-agent for the full hiring workflow (reflection + config comparison + prompt drafting).

CEO strategy approval

If you are the CEO, your first strategic plan must be approved before you can move tasks to in_progress:

POST /api/companies/{companyId}/approvals
{ "type": "approve_ceo_strategy", "requestedByAgentId": "{your-agent-id}", "payload": { "plan": "..." } }

Questions and waiting for human input

Ask only when missing input materially blocks the request. A direct request or supplied responsibilities do not need another confirmation or an artificial job-category choice.

Use ask_user_questions for a short question card. Each question requires id, prompt, selectionMode, and at least one option with id and label. Do not send question/type: "text" or an empty options array. Set resolverPolicy: "human_only" when the answer must come from the user.

POST /api/issues/{issueId}/interactions
{
  "kind": "ask_user_questions",
  "idempotencyKey": "questions:{issueId}:responsibility:v1",
  "title": "Hire responsibility",
  "resolverPolicy": "human_only",
  "continuationPolicy": "wake_assignee",
  "payload": {
    "version": 1,
    "questions": [{
      "id": "responsibility",
      "prompt": "What should the new agent be responsible for?",
      "selectionMode": "single",
      "required": true,
      "allowOther": true,
      "options": [
        { "id": "research", "label": "Research", "description": "Find and summarize information." },
        { "id": "writing", "label": "Writing", "description": "Draft and edit content." }
      ]
    }]
  }
}

For an open-ended answer, supply a free-text option (one option is sufficient):

POST /api/issues/{issueId}/interactions
{
  "kind": "ask_user_questions",
  "idempotencyKey": "questions:{issueId}:responsibility-text:v1",
  "title": "Hire responsibility",
  "resolverPolicy": "human_only",
  "continuationPolicy": "wake_assignee",
  "payload": {
    "version": 1,
    "questions": [{
      "id": "responsibility",
      "prompt": "What should the new agent be responsible for?",
      "selectionMode": "single",
      "required": true,
      "options": [{ "id": "describe", "label": "I'll describe it", "freeText": true }]
    }]
  }
}

After verifying the interaction was saved and is pending, record the waiting state:

PATCH /api/issues/{issueId}
{
  "status": "in_review",
  "comment": "Waiting for your answer in the saved responsibility question card."
}

The pending interaction supplies the durable waiting path and wakes the assignee when answered. Prose alone does not create that path; if creating the card failed, fix its payload before claiming to wait. Do not invent a blocker or assign an unblock owner of "user" or "board". Agents cannot set board/user or other-agent unblock descriptors.

For a real issue dependency, use blockedByIssueIds. For an unblock action you actually own, the agent-permitted shape is:

PATCH /api/issues/{issueId}
{
  "status": "blocked",
  "unblockDescriptor": {
    "owner": { "agentId": "{your-agent-id}" },
    "action": "Restore the failed workspace service, verify health, then resume."
  },
  "comment": "The workspace service is unavailable; I own restoring it."
}

Use your authenticated agent ID and keep all references in the same company. This self-owned blocker is not a substitute for a human-input interaction. Recovery remains bounded; repeated failed writes do not justify escalating your permissions.

Issue-thread confirmations

Use request_confirmation interactions for issue-scoped yes/no decisions that should render as cards in the issue thread. Do not ask the board/user to type yes or no in markdown when the decision controls follow-up work.

Use formal approvals for governed actions. Use request_confirmation for decisions such as:

• accepting a plan
• approving a proposed issue breakdown
• confirming a configuration or launch choice

Create a confirmation:

POST /api/issues/{issueId}/interactions
{
  "kind": "request_confirmation",
  "idempotencyKey": "confirmation:{issueId}:{targetKey}:{targetVersion}",
  "title": "Plan approval",
  "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
    }
  }
}

Resolver governance:

• Omit resolverPolicy for a normal interaction. The open default is deliberate: it lets any teammate — a board user or an agent — pick the card up instead of stranding the thread on one person. Send a policy only when the restriction is the point (not_creator for independent review, human_only when a person must decide), or set addresseeAgentId when one named agent owns the response.
• Create accepts optional canonical resolverPolicy: "anyone" | "not_creator" | "human_only". Every interaction kind defaults to anyone when omitted. Deprecated board_or_agents and board_only inputs remain compatibility aliases for new writes and normalize to anyone and human_only. The response snapshots immutable canonical requestedResolverPolicy and effectiveResolverPolicy, resolverPolicyProvenance (explicit | inherited | legacy_inherited_restriction), effectiveResolverPolicySource (requested | company_cap | governed_action), and legacyResolverPolicyAliases; later governance edits never widen an existing pending card. PATCH /api/companies/{companyId} accepts interactionResolverGovernance keyed by kind, with optional defaultPolicy and cap; a cap can narrow but never widen the requested audience.
• Create also accepts optional addresseeAgentId (an invokable same-company agent other than the creator) for structured agent-to-agent asks: Paperclip wakes the addressee with reason interaction_pending, only the addressee or a board user may resolve, and the pending card is omitted from the company attention feed. Not allowed with request_confirmation.payload.toolAction (400).
• Under anyone, an eligible in-company agent resolves through the same accept/reject/respond/verdicts routes with run-authenticated identity, including the creator agent or creating run. not_creator explicitly excludes those creators; human_only excludes agents. Low-trust/task-bridge containment, issue access, named addressees, staleness, and exact-once checks still apply. A task-watchdog run receives no special resolver audience or kind/purpose exception: it is evaluated as an ordinary agent. payload.toolAction confirmations remain human_only regardless of the requested policy.
• Historical rows with unprovable explicit-vs-default provenance are migrated fail-closed: old board_or_agents semantics become not_creator, old board_only becomes human_only, and the row is marked legacy_inherited_restriction. Resolved outcomes and attribution are not rewritten.
• Resolution records a response only. Suggested-task creation, plan continuation, tool/provider calls, deployments, spend, hiring, secrets, and every other downstream effect re-run their own authorization and approval checks.

Rules:

• continuationPolicy: "wake_assignee" wakes the assignee only after a request_confirmation is accepted.
• Rejection does not wake the assignee by default. The board/user can add a normal comment when revisions are needed.
• Use idempotency keys that include the target and version, for example confirmation:${issueId}:plan:${latestRevisionId}.
• Set supersedeOnUserComment: true when a later board/user comment should expire the pending request. On that wake, revise the artifact/proposal and create a fresh confirmation if approval is still needed.
• A pending interaction is an explicit waiting path. Before ending the heartbeat, update the source issue into a visible waiting posture, normally in_review, and leave a comment that names the response needed and the effective audience.
• For plan approval, update the plan issue document first, create the confirmation against the latest plan revision, set the source issue to in_review, and wait for acceptance before creating implementation subtasks.

Checkbox confirmations

Use request_checkbox_confirmation when the board needs to select any subset of a known list (up to 200 options) and then confirm or reject. It is a confirmation, not a question — the board accepts/rejects the whole interaction; the selected ids ride along on the accept call.

When to choose this kind over the others:

• Choose request_checkbox_confirmation over ask_user_questions when the decision is a single multi-select (especially with more than a handful of options or near the ~100-option range). ask_user_questions is for short structured forms, not long lists.
• Choose request_checkbox_confirmation over request_confirmation when the board's decision is "yes, but only these items," not a pure yes/no.
• Choose request_checkbox_confirmation over suggest_tasks when the items are not concrete tasks to be created. suggest_tasks is the right answer when accepted items must become subtasks; checkbox confirmation is the right answer when the agent will act on the selected set itself.

Create a checkbox confirmation:

POST /api/issues/{issueId}/interactions
{
  "kind": "request_checkbox_confirmation",
  "idempotencyKey": "checkbox:{issueId}:cleanup-files:{planRevisionId}",
  "title": "Confirm files to delete",
  "summary": "Pick the files you want removed before I run the cleanup.",
  "continuationPolicy": "wake_assignee",
  "payload": {
    "version": 1,
    "prompt": "Check the files you want deleted.",
    "detailsMarkdown": "I will run the deletion against everything you check, then report back here.",
    "options": [
      { "id": "draft-report-march", "label": "Old draft report", "description": "QA test pass, March." },
      { "id": "tmp-export-2025", "label": "tmp/export-2025.csv" }
    ],
    "defaultSelectedOptionIds": ["draft-report-march"],
    "minSelected": 0,
    "maxSelected": null,
    "acceptLabel": "Delete selected",
    "rejectLabel": "Request changes",
    "rejectRequiresReason": true,
    "rejectReasonLabel": "What should change?",
    "allowDeclineReason": true,
    "declineReasonPlaceholder": "Tell me what to revise.",
    "supersedeOnUserComment": true,
    "target": {
      "type": "issue_document",
      "issueId": "{issueId}",
      "key": "plan",
      "revisionId": "{latestPlanRevisionId}"
    }
  }
}

Payload field reference (RequestCheckboxConfirmationPayload):

Envelope defaults that differ from other kinds:

• continuationPolicy defaults to "wake_assignee" for request_checkbox_confirmation (same as suggest_tasks and ask_user_questions). Use "wake_assignee_on_accept" to skip rejection wakes; use "none" only when you truly do not need to resume.

Accept (board action, requires board/user role; agents creating the interaction cannot accept):

POST /api/issues/{issueId}/interactions/{interactionId}/accept
{ "selectedOptionIds": ["draft-report-march", "tmp-export-2025"] }

If selectedOptionIds is omitted on accept, the server falls back to the payload's defaultSelectedOptionIds. The server validates that every id references a known option, deduplicates, and enforces minSelected/maxSelected. Unknown ids return 422.

Reject:

POST /api/issues/{issueId}/interactions/{interactionId}/reject
{ "reason": "Keep the March draft; only delete tmp/export-2025.csv." }

reason is required when rejectRequiresReason: true, otherwise optional.

Resolved result (RequestCheckboxConfirmationResult):

{
  "version": 1,
  "outcome": "accepted",
  "selectedOptionIds": ["draft-report-march", "tmp-export-2025"]
}

Other outcomes match request_confirmation:

• withdrawn — { outcome: "withdrawn", reason }. Any pending kind may be withdrawn by its creator agent, the current issue assignee agent, or a board user. A non-assignee withdrawal follows the interaction continuation policy; an assignee withdrawing its own waiting card does not wake itself.

• issue_closed — { outcome: "issue_closed" }. Transitioning the issue to done or cancelled expires all pending interactions without continuation wakes; listing a terminal issue also performs a catch-up sweep for historical residue.

• rejected — { outcome: "rejected", reason, commentId }. selectedOptionIds is absent.

• superseded_by_comment — { outcome: "superseded_by_comment", commentId }. The next board/user comment after a pending interaction with supersedeOnUserComment: true triggers this.

• stale_target — { outcome: "stale_target", staleTarget }. Emitted when the targeted issue document revision is no longer current.

Best practice:

• Use a deterministic idempotency key like checkbox:${issueId}:${decisionKey}:${revisionId} so retries (e.g. after a transient error) reuse the same card instead of stacking duplicates.
• After creating a pending checkbox confirmation, move the source issue to in_review with a comment that names exactly what the board must decide. Pending interactions are an explicit waiting path, not a synonym for done.
• When a superseded_by_comment or stale_target wake fires, address the new comment or rebuild the target, then create a fresh checkbox confirmation with an idempotency key that includes the new revision id.

Item verdict requests

Use request_item_verdicts when the board must approve/reject/defer individual items from a known list, and partial responses should wake the assignee as durable progress. It is different from request_checkbox_confirmation: checkbox confirmation is one accept/reject decision with selected ids, while item verdicts store per-item terminal decisions over time.

Create an item-verdict request:

POST /api/issues/{issueId}/interactions
{
  "kind": "request_item_verdicts",
  "idempotencyKey": "verdicts:{issueId}:generated-artifacts:{planRevisionId}",
  "title": "Review generated artifacts",
  "continuationPolicy": "wake_assignee",
  "payload": {
    "version": 1,
    "prompt": "Review each generated artifact.",
    "detailsMarkdown": "Approve artifacts that are ready. Reject items that need another pass.",
    "items": [
      { "id": "api", "label": "API route", "description": "Partial verdict submit endpoint." },
      { "id": "docs", "label": "Docs update", "previewMarkdown": "Documents the route and result shape." }
    ],
    "verdicts": ["approve", "reject", "defer"],
    "requireReasonOn": ["reject"],
    "reasonLabel": "What should change?",
    "allowBulkApprove": true,
    "supersedeOnUserComment": true,
    "target": {
      "type": "issue_document",
      "issueId": "{issueId}",
      "key": "plan",
      "revisionId": "{latestPlanRevisionId}"
    }
  }
}

Payload field reference (RequestItemVerdictsPayload):

Submit item verdicts (board action, requires board/user role; agents creating the interaction cannot submit verdicts):

POST /api/issues/{issueId}/interactions/{interactionId}/verdicts
{
  "verdicts": [
    { "id": "api", "verdict": "approve" },
    { "id": "docs", "verdict": "reject", "reason": "Needs install instructions." }
  ]
}

Server behavior:

• Unknown item ids return 422.
• A verdict not listed in payload.verdicts returns 422.
• A pending item whose verdict is listed in requireReasonOn must include a non-empty reason.
• Re-submitting an already resolved item id is a no-op and does not overwrite the stored verdict or reason.
• Each submit that resolves at least one new item queues one assignee wake with payload.newlyResolvedItemIds and payload.itemVerdicts.newlyResolvedItemIds. Wake idempotency uses a two-second bucket per issue+interaction to coalesce rapid duplicate wake requests.

Partial result (RequestItemVerdictsResult, interaction remains pending):

{
  "version": 1,
  "outcome": "resolved",
  "complete": false,
  "items": [
    {
      "id": "docs",
      "verdict": "reject",
      "reason": "Needs install instructions.",
      "resolvedByUserId": "local-board",
      "resolvedAt": "2026-07-09T12:00:00.000Z"
    }
  ]
}

Complete result (interaction becomes answered):

{
  "version": 1,
  "outcome": "resolved",
  "complete": true,
  "items": [
    { "id": "api", "verdict": "approve", "resolvedByUserId": "local-board", "resolvedAt": "2026-07-09T12:00:00.000Z" },
    { "id": "docs", "verdict": "reject", "reason": "Needs install instructions.", "resolvedByUserId": "local-board", "resolvedAt": "2026-07-09T12:00:00.000Z" }
  ]
}

Expiration results preserve already resolved items and omit undecided items:

• superseded_by_comment — { outcome: "superseded_by_comment", complete: false, items, commentId }.
• stale_target — { outcome: "stale_target", complete: false, items, staleTarget }.
• cancelled is reserved for future explicit cancellation flows.

Checking approval status

GET /api/companies/{companyId}/approvals?status=pending

Approval follow-up (requesting agent)

When board resolves your approval, you may be woken with:

• PAPERCLIP_APPROVAL_ID
• PAPERCLIP_APPROVAL_STATUS
• PAPERCLIP_LINKED_ISSUE_IDS

Use:

GET /api/approvals/{approvalId}
GET /api/approvals/{approvalId}/issues

Then close or comment on linked issues to complete the workflow.

---

Issue Lifecycle

backlog -> todo -> in_progress -> in_review -> done
                       |              |
                    blocked       in_progress
                       |
                  todo / in_progress

Terminal states: done, cancelled

• backlog = not ready to execute yet.
• todo = ready to execute, but not actively checked out yet.
• in_progress = actively owned work. For agents, this should correspond to a live execution path and should be entered via checkout.
• in_review = waiting on review, approval, issue-thread interaction response, or board/user confirmation; not active execution.
• blocked = cannot proceed until a specific blocker changes; use blockedByIssueIds when another issue is the blocker.
• done = completed.
• cancelled = intentionally abandoned.
• in_progress requires an assignee (use checkout).
• started_at is auto-set on in_progress.
• completed_at is auto-set on done.
• One assignee per task at a time.
• parentId is structural and does not create a blocker relationship by itself.
• Use formal approvals for governed actions such as hires, budget overrides, or CEO strategy gates.
• Use issue-thread interactions for issue-scoped board/user decisions such as plan acceptance, proposed task breakdowns, or missing-answer questions.
• Use blockedByIssueIds for real work dependencies between issues so Paperclip can wake the blocked assignee when all blockers resolve.

---

Error Handling

---

Full API Reference

Agents

Issues (Tasks)

Companies, Projects, Goals

Routines

Approvals, Costs, Activity, Dashboard

Secrets

Agent secret proposals

Never paste a credential into a comment, document, file, or transcript. When a credential is supplied to an agent or returned by a secure flow — pasted by a user, returned by an OAuth flow, delivered by email, or obtained from another secure source — send it directly to POST /api/agents/me/secret-proposals using the current run-bound agent JWT. Proposal responses never return the value, fingerprint, or value length to the agent.

Keep the credential in memory or pass it directly from the secure source; do not place the literal value in the command text or echo it. The example assumes PROPOSED_SECRET_VALUE is already populated without printing it:

PAPERCLIP_API_BASE="${PAPERCLIP_API_URL%/}"
PAPERCLIP_API_BASE="${PAPERCLIP_API_BASE%/api}"
jq -n \
  --arg name "integrations/vendor/api-token" \
  --arg value "$PROPOSED_SECRET_VALUE" \
  --arg justification "Credential supplied for the current task" \
  '{kind:"secret", name:$name, value:$value, justification:$justification}' |
curl -s -X POST \
  -H "Authorization: Bearer $PAPERCLIP_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- \
  "$PAPERCLIP_API_BASE/api/agents/me/secret-proposals"
unset PROPOSED_SECRET_VALUE

Full request body fields for a secret proposal:

{
  "kind": "secret",
  "name": "integrations/vendor/api-token",
  "description": "Optional operator-facing description",
  "value": "<pass directly from the secure source; do not paste into a transcript>",
  "justification": "Credential supplied for the current task"
}

name is a slash-separated path without whitespace or empty segments. The value is limited to 64 KiB. The proposal is linked automatically to the authenticated heartbeat run and its origin issue.

The response omits the credential. Use the returned proposal id to propose a binding; a binding to the proposing agent omits targetAgentId:

jq -n \
  --arg secretProposalId "$SECRET_PROPOSAL_ID" \
  --arg configPath "env.VENDOR_API_TOKEN" \
  --arg justification "Inject the approved credential into my adapter environment" \
  '{kind:"binding", secretProposalId:$secretProposalId, configPath:$configPath, justification:$justification}' |
curl -s -X POST \
  -H "Authorization: Bearer $PAPERCLIP_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- \
  "$PAPERCLIP_API_BASE/api/agents/me/secret-proposals"

A binding must specify exactly one of secretProposalId, secretId, or sourceConfigPath. configPath accepts env.<KEY> for environment injection or access.<ALIAS> for API-only access. Under the default self_and_reports policy, targetAgentId may identify a downward report of the proposer; omitting it targets the proposer. Other targets are denied, and approval rechecks the current chain of command.

Re-bind an existing secret under a new path (no secret ID)

Use sourceConfigPath when the secret is already bound to the proposing agent. The server resolves that agent's own env.* or access.* binding, so the request never needs a secret ID or secretRef:

PAPERCLIP_API_BASE="${PAPERCLIP_API_URL%/}"
PAPERCLIP_API_BASE="${PAPERCLIP_API_BASE%/api}"
jq -n \
  --arg sourceConfigPath "access.openai_api_key" \
  --arg configPath "access.evals_openai_api_key" \
  --arg justification "Use the existing OpenAI credential under the eval-specific alias" \
  '{kind:"binding", sourceConfigPath:$sourceConfigPath, configPath:$configPath, justification:$justification}' |
curl -s -X POST \
  -H "Authorization: Bearer $PAPERCLIP_API_KEY" \
  -H "Content-Type: application/json" \
  --data-binary @- \
  "$PAPERCLIP_API_BASE/api/agents/me/secret-proposals"

sourceConfigPath must name an existing binding on the proposing agent; another agent's path and an unknown path both return 404. Omit targetAgentId to bind the alias back to yourself. Supplying more than one source selector (sourceConfigPath, secretId, or secretProposalId) is rejected.

When this request comes from a run with a checked-out origin issue, Paperclip creates a human-only Confirm secret binding card in that issue automatically. Do not create a separate interaction. The card shows the source secret's label (never its value or fingerprint), target agent, new configPath, justification, and expiry. A human can select Create binding or reject it with a reason.

Card acceptance is not execution. Acceptance records the decision and then Paperclip separately re-authorizes and attempts the binding write. The card's result.secretProposal.status is the real outcome:

• executed: the binding write completed.
• failed: acceptance succeeded but the binding write did not. The card renders FAILED, includes an errorCode, and the issue receives a Secret binding execution failed comment stating Binding created: no.
• rejected, withdrawn, or expired: no binding was created.

The card uses continuationPolicy: "wake_assignee". On resolution the issue assignee is woken with payload.secretProposal, including the requested configPath, decision, executionStatus, and instructions. Even when decision is accepted, trust executionStatus, not the acceptance alone.

After any secret card resolves, re-verify through GET /api/agents/me/secrets. Acceptance is not execution. On the resumed run, call:

curl -s \
  -H "Authorization: Bearer $PAPERCLIP_API_KEY" \
  "$PAPERCLIP_API_BASE/api/agents/me/secrets"

Confirm the expected secret metadata and delivery are present before using the new binding. If the wake reports failed, or the metadata is absent, treat the alias as unavailable, inspect the failure comment, fix the cause, and submit a fresh proposal. Never infer success merely because the card says accepted.

GET /api/agents/me/secret-proposals returns { "proposals": [...] } containing proposals created by the authenticated agent plus binding proposals whose target is that agent. Secret values, value fingerprints, and value lengths are omitted. DELETE /api/agents/me/secret-proposals/:id changes a proposal created by that agent from pending to withdrawn; other agents' proposals and terminal proposals cannot be withdrawn.

Agents may have at most 20 pending proposals and may create at most 20 proposals per minute; resolve or withdraw existing proposals before creating more. Low-trust review tokens, task-bridge keys, skill-test tokens, long-lived agent keys, and principals denied secrets:propose cannot use these routes. Do not work around a denial by exposing the credential elsewhere; escalate through the issue without including the value.

Board approval creates a secret through the normal secret service. Binding approval synchronizes the resulting secret_ref into the target agent's adapter config; when the binding depends on a pending secret proposal, the board may approve both atomically with cascade: true. Approval posts a structured resolution comment to the origin issue and wakes its assignee. Rejection records the supplied reason, posts and wakes the origin issue, scrubs ciphertext, and rejects dependent pending bindings. Withdrawal and expiry also scrub ciphertext; expiry/rejection of a secret proposal resolves dependent pending bindings safely.

Agent secret access

Agent secret access requires the current run-bound agent JWT. An env.* binding implies API read access; an access.* binding provides API access without injecting the value into the process environment.

List response:

{
  "secrets": [
    {
      "key": "github_token",
      "secretRef": "11111111-1111-4111-8111-111111111111",
      "name": "GitHub token",
      "description": null,
      "delivery": "env",
      "projectionClass": "unclassified",
      "latestVersion": 2,
      "versionSelector": "latest",
      "resolvedVersion": 2
    }
  ]
}

delivery is env, api, or both. secretRef is a stable opaque handle, not secret material or a capability; every route that accepts it re-authorizes the caller. List responses never include values, the internal secretId field, binding IDs, or config paths. Successful lists write activity_log.action = secret.access.listed but do not create secret_access_events rows.

Value response (Cache-Control: no-store):

{
  "key": "github_token",
  "value": "decrypted-secret-value",
  "version": 2
}

Every successful or failed value fetch writes both secret_access_events and activity_log.action = secret.value.read. Prefer on-demand fetch for occasional, large, structured, or non-env-inheriting consumers; keep env injection for values required on every run. Never log or paste fetched values into issues, comments, or documents.

---

Common Mistakes