---
title: "Fork a workspace into a new key (independent copy of its committed state)"
method: POST
path: "/v1/workspaces/{workspace_id}/fork"
tags: ["Workspaces"]
---

# Fork a workspace into a new key (independent copy of its committed state)

`POST /v1/workspaces/{workspace_id}/fork`

## Request

- Base URL: `https://api.shardflux.dev`
- URL: `https://api.shardflux.dev/v1/workspaces/{workspace_id}/fork`
- Auth: HTTP bearer

Creates the target workspace (same template version and disk layout) and a `fork` operation on it, admitted like a start. Caps default to the source’s. `lifetime` is the fork’s own (default persistent): forking a session keeps its state. A file-first source is 409 not_supported_for_mode. Supports Idempotency-Key. Caps (and the memory allocation mode) default to the source’s. Held fork: with Prefer: wait=<seconds> (at most 20), holds through the terminal operation. Success returns 200 with the running target, succeeded operation and a final-epoch tool token for agent_label/tools. Otherwise returns 202 with the fresh operation. Preference-Applied confirms the wait; without it, poll as usual. A token that could not be issued is null; fetch it from the target tool-token route. Errors: 402/403/429 as for open (admitted like a start). Plan concurrency: a start the plan’s concurrent_workspaces limit refuses is queued on the request (up to 10 s) and admitted as soon as one of the organization’s workspaces frees; if none does, 429 `quota_exceeded` with `Retry-After: 2` (retryable; details limit concurrent_workspaces, limit_value, current, retry_after_seconds, plan). Nothing was created: retry after Retry-After, with the same Idempotency-Key or a new one. With `Prefer: wait` the queued time counts against the wait. Memory allocation: caps.allocation_mode elastic with caps.memory_mib_held; 422 validation_failed details.field caps.allocation_mode reason allocation_mode_not_available (the organization lacks elastic memory; nothing is created or changed), reason not_supported_for_mode (file_first); details.field caps.memory_mib_held reason requires_elastic (held without elastic) or exceeds_memory_mib (held above caps.memory_mib).

## Path parameters

- `workspace_id` string, uuid, required

## Headers

- `prefer` string

## Request body

- object
  - `key` string, required: Stable workspace key, unique per organization (e.g. `${customerId}/${projectId}`).
  - `caps` object: Optional user caps; the ceiling is min(template, cap, plan). Absent fields add no restriction. Given caps replace the stored ones (a reopen keeps an elastic workspace’s mode when elastic is the organization’s default and allocation_mode is omitted).
    - `cpu_millis` integer
    - `memory_mib` integer: Fixed: size in MiB; omitted uses the template default (template defaults.memory_mib, else manifest default_caps.memory_mib_base, else 1024), bounded by plan and template. Elastic: maximum memory promise.
    - `disk_gib` integer: Disk (GiB) for the workspace's own changes, bounded by plan and template. A new workspace without it starts at the template's default disk (manifest default_caps.disk_gib, 20 GiB on the platform templates), or, opened from your saved machine or project template, at the disk of the workspaces it was saved from when that is larger; grow it later with a resize. A reopen without it keeps the disk.
    - `allocation_mode` 'fixed' | 'elastic': Memory allocation: fixed boots memory_mib (the template default when omitted) and holds it; elastic makes memory_mib a promise: the VM holds memory_mib_held while idle and is grown towards the promise when a command needs it. Needs the elastic memory entitlement (422 details.reason allocation_mode_not_available). A promise that does not exceed the held floor by at least 512 MiB is fixed at the promise (the response caps say which). Omitted: fixed, or elastic where that is the organization’s default (a reopen then keeps the stored mode). Takes effect at the next VM start.
    - `memory_mib_held` integer: Elastic only: memory held while idle (MiB, default 1024, at most memory_mib).
  - `lifetime` 'persistent' | 'session': persistent: kept until deleted (default). session: discarded when the session ends (close(), idle timeout or draft discard;).
  - `labels` object: Searchable workspace metadata; not secrets. At most 50 labels.
  - `agent_label` string: Attribution label; one agent session per (workspace, principal, label).
  - `tools` string[]

## Response `200`

Default Response

- object
  - `operation` Operation, required
    - `id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `kind` 'open' | 'suspend' | 'resume' | 'fork' | 'snapshot' | 'restore' | 'delete' | 'move' | 'resize' | 'upgrade' | 'volume_create' | 'volume_attach' | 'volume_detach' | 'volume_delete' | 'reset' | 'layer_snapshot', required
    - `state` 'queued' | 'capacity_pending' | 'running' | 'succeeded' | 'failed' | 'canceled', required
    - `workspace_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
    - `volume_id` string, uuid, nullable: UUIDv7, lowercase canonical form.
    - `created_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `updated_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `started_at` string, date-time, nullable: RFC 3339 UTC timestamp with Z.
    - `completed_at` string, date-time, nullable: RFC 3339 UTC timestamp with Z.
    - `state_reason` string, nullable: Closed reason code while waiting (e.g. capacity_pending: no_ready_host).
    - `input` object
    - `error` object
    - `result` object
      - `durable` boolean: suspend and fork: whether the capture is in durable storage. false while `durability.state` is pending (the suspend completed on its host) or lost; true once it committed. Absent on other kinds.
      - `suspend_path` string: suspend: `local_commit` when the suspend completed on its host and the durable copy followed in the background.
      - `durability` object
        - `state` 'pending' | 'durable' | 'lost', required: pending: sealed on its host, the durable copy is being written; durable: the copy committed; lost: the copy could not be made and the previous recovery point stays current.
        - `checkpoint_id` string: The checkpoint being made durable.
        - `generation_id` string: Its generation on the host.
        - `local_commit_at` string: RFC 3339: when the suspend (or fork) completed on its host.
        - `durable_by` string: RFC 3339: when the durable copy is expected to have committed (local_commit_at + the deadline).
        - `durable_at` string: RFC 3339 (state durable): when the durable copy committed.
        - `local_commit_to_durable_ms` number: State durable: local_commit_at to durable_at, in milliseconds.
        - `overdue_at` string: RFC 3339: set once durable_by passed while the copy was still pending; it keeps retrying.
        - `reason` string: State lost: why, e.g. host_lost.
        - `code` string: State lost: the error code of the failed copy.
        - `recovery_point_checkpoint_id` string, nullable: State lost: the durable checkpoint that stays the recovery point (null when there is none).
      - `lost_suspend` object
        - `checkpoint_id` string, required: The suspend checkpoint whose durable copy was lost.
        - `generation_id` string: Its generation on the host.
        - `reason` string: Why it was lost, e.g. host_lost.
        - `suspended_at` string: RFC 3339: when that suspend captured the workspace.
        - `restored_checkpoint_id` string: The earlier durable checkpoint this resume restored.
        - `state_as_of` string: RFC 3339: when the restored checkpoint committed; the workspace state is as of then.
      - `host_lost` object
        - `detected_at` string, required: RFC 3339: when the host failure was detected; the workspace was suspended then.
        - `restored_from` 'disk' | 'checkpoint': resume and open: disk when the workspace booted from its disk (files kept, processes and memory not); checkpoint when it was restored from its newest checkpoint (see state_as_of). Absent on a suspend.
        - `restored_checkpoint_id` string: restored_from checkpoint: the checkpoint this resume restored.
        - `state_as_of` string: restored_from checkpoint, RFC 3339: when that checkpoint committed; the workspace state is as of then.
    - `progress` SuspendProgress
      - `operation_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `state` 'queued' | 'capacity_pending' | 'running' | 'succeeded' | 'failed' | 'canceled', required
      - `workspace_state` 'creating' | 'starting' | 'running' | 'suspending' | 'suspended' | 'resuming' | 'forking' | 'stopping' | 'failed' | 'deleting' | 'deleted', required
      - `phase` string, required: Actual snapshot preparation/capture phase, or an explicitly unavailable control status.
      - `cancelable` boolean, required: True only while the host can acknowledge cancellation before its capture cut.
      - `supported` boolean, required
      - `safe_action` 'wait_for_terminal' | 'resume' | 'retry_tools' | 'inspect_failure', required
      - `phase_started_at` string, date-time: RFC 3339 UTC timestamp with Z.
      - `deadline_at` string, date-time: RFC 3339 UTC timestamp with Z.
      - `accepted` boolean
  - `workspace` Workspace, required
    - `secret_redaction_required` boolean: Bound secrets require tool-call capture redaction inside the guest before storage.
    - `agents` Agent[]
      - `id` string, required
      - `workspace_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `workspace_key` string, required
      - `app` union, required
        - 'claude'
        - 'codex'
      - `state` union, required
        - 'working'
        - 'waiting'
        - 'exited'
      - `since` string, date-time, required: RFC 3339 UTC timestamp with Z.
      - `title` string, required
      - `remote_control_url` string, required
      - `last_message` string, required: The session's latest message as one line of plain text: the agent's last reply when it waits or exits, its latest text while it works. Empty until it says one.
      - `labels` object, required
      - `conversation_id` string, nullable: The session's own conversation: the one whose harness session it is, else its computer's only live conversation. Several threads can share one computer, so prefer it over labels.shard_conversation, which names the computer's first conversation. Null when it cannot be told; absent before conversations exist.
      - `orchestrator_conversation_id` string, nullable
      - `orchestrator_input_id` string, nullable
      - `settled_reason` union: The conversation computer suspended or failed while a native turn was active. Null or absent does not establish completion.
        - 'computer_suspended'
        - 'computer_failed'
      - `credential_status` union
        - 'ready'
        - 'needs_sign_in'
    - `id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `organization_id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `project_id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `workspace_key` string, required
    - `labels` object, required: Searchable workspace metadata; not secrets. At most 50 labels.
    - `retention` object, nullable
      - `delete_after_idle_days` integer, required
      - `delete_at` string, date-time, nullable, required
    - `template_version_id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `template` object, required: The workspace’s current template base (changed only by an explicit upgrade), and the version its immutable paths show.
      - `version_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `template_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `slug` string, required
      - `version` integer, required
      - `immutable_version` integer, nullable, required: Immutable paths: the template version whose read-only content the workspace has mounted at the template’s immutable paths. It follows the template’s newest published version at a cold boot or a resume. null when the workspace mounts none (its version declares no immutable paths, a draft or a test instance, or not booted yet).
    - `desired_state` 'running' | 'suspended' | 'deleted', required
    - `observed_state` 'creating' | 'starting' | 'running' | 'suspending' | 'suspended' | 'resuming' | 'forking' | 'stopping' | 'failed' | 'deleting' | 'deleted', required
    - `cell_id` string, nullable, required
    - `cell_endpoint` string, nullable, required: Cell gateway base URL when placed and resolvable.
    - `ownership_epoch` integer, required
    - `caps` object, required: CPU and memory user caps (null = no user restriction), the effective disk cap, and the memory allocation mode.
      - `cpu_millis` integer, nullable, required
      - `memory_mib` integer, nullable, required
      - `disk_gib` integer, nullable, required: Requested disk cap (GiB), bounded by the admitted plan/template ceiling and acknowledged VM disk. Null means no user disk restriction.
      - `allocation_mode` 'fixed' | 'elastic', required: Memory allocation: fixed (default) boots memory_mib and holds it; elastic holds memory_mib_held while idle and is grown up to memory_mib (the promise) when a command needs it. The mode the next VM start uses: an elastic request whose promise does not exceed the held floor by at least 512 MiB is fixed at the promise.
      - `memory_mib_held` integer: Elastic only: the memory held while idle (MiB).
    - `memory` object, required: Memory of the workspace.
      - `allocation_mode` 'fixed' | 'elastic', required: The layout of the running VM, else the one the next start uses.
      - `promised_mib` integer, nullable, required: The memory the workspace may use (MiB): the running allocation, else the last admitted size; null before the first start.
      - `held_mib` integer, nullable, required: The memory held while idle (MiB): the elastic floor; equal to promised_mib for fixed.
      - `plugged_mib` integer, nullable, required: Elastic and running: the memory currently plugged above the held floor (MiB, from the host’s latest report); null otherwise.
    - `ceilings` object, nullable, required: Effective per-workspace ceilings computed at the last admission (start/resume/fork).
      - `cpu_millis` integer, required
      - `memory_mib` integer, required
      - `disk_gib` integer, required
      - `sources` object, required: Which input bounded each ceiling: min(template, user cap, plan cap).
        - `cpu_millis` 'template' | 'user' | 'plan', required
        - `memory_mib` 'template' | 'user' | 'plan', required
        - `disk_gib` 'template' | 'user' | 'plan', required
      - `resolved_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `grants` object, nullable, required: Actual grants reported by the cell; live memory allocation overrides the last successful start.
    - `storage_recovery` object: Files stored independently of compute. Resume restarts processes from the kept filesystem.
      - `state` 'stored', required
      - `reason` 'maintenance', required
      - `copy_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `files_kept` true, required
      - `message` string, required
    - `active_operation` Operation, required
      - `id` string, uuid, required: UUIDv7, lowercase canonical form.
      - `kind` 'open' | 'suspend' | 'resume' | 'fork' | 'snapshot' | 'restore' | 'delete' | 'move' | 'resize' | 'upgrade' | 'volume_create' | 'volume_attach' | 'volume_detach' | 'volume_delete' | 'reset' | 'layer_snapshot', required
      - `state` 'queued' | 'capacity_pending' | 'running' | 'succeeded' | 'failed' | 'canceled', required
      - `workspace_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
      - `volume_id` string, uuid, nullable: UUIDv7, lowercase canonical form.
      - `created_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
      - `updated_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
      - `started_at` string, date-time, nullable: RFC 3339 UTC timestamp with Z.
      - `completed_at` string, date-time, nullable: RFC 3339 UTC timestamp with Z.
      - `state_reason` string, nullable: Closed reason code while waiting (e.g. capacity_pending: no_ready_host).
      - `input` object
      - `error` object
      - `result` object
        - `durable` boolean: suspend and fork: whether the capture is in durable storage. false while `durability.state` is pending (the suspend completed on its host) or lost; true once it committed. Absent on other kinds.
        - `suspend_path` string: suspend: `local_commit` when the suspend completed on its host and the durable copy followed in the background.
        - `durability` object
          - `state` 'pending' | 'durable' | 'lost', required: pending: sealed on its host, the durable copy is being written; durable: the copy committed; lost: the copy could not be made and the previous recovery point stays current.
          - `checkpoint_id` string: The checkpoint being made durable.
          - `generation_id` string: Its generation on the host.
          - `local_commit_at` string: RFC 3339: when the suspend (or fork) completed on its host.
          - `durable_by` string: RFC 3339: when the durable copy is expected to have committed (local_commit_at + the deadline).
          - `durable_at` string: RFC 3339 (state durable): when the durable copy committed.
          - `local_commit_to_durable_ms` number: State durable: local_commit_at to durable_at, in milliseconds.
          - `overdue_at` string: RFC 3339: set once durable_by passed while the copy was still pending; it keeps retrying.
          - `reason` string: State lost: why, e.g. host_lost.
          - `code` string: State lost: the error code of the failed copy.
          - `recovery_point_checkpoint_id` string, nullable: State lost: the durable checkpoint that stays the recovery point (null when there is none).
        - `lost_suspend` object
          - `checkpoint_id` string, required: The suspend checkpoint whose durable copy was lost.
          - `generation_id` string: Its generation on the host.
          - `reason` string: Why it was lost, e.g. host_lost.
          - `suspended_at` string: RFC 3339: when that suspend captured the workspace.
          - `restored_checkpoint_id` string: The earlier durable checkpoint this resume restored.
          - `state_as_of` string: RFC 3339: when the restored checkpoint committed; the workspace state is as of then.
        - `host_lost` object
          - `detected_at` string, required: RFC 3339: when the host failure was detected; the workspace was suspended then.
          - `restored_from` 'disk' | 'checkpoint': resume and open: disk when the workspace booted from its disk (files kept, processes and memory not); checkpoint when it was restored from its newest checkpoint (see state_as_of). Absent on a suspend.
          - `restored_checkpoint_id` string: restored_from checkpoint: the checkpoint this resume restored.
          - `state_as_of` string: restored_from checkpoint, RFC 3339: when that checkpoint committed; the workspace state is as of then.
      - `progress` SuspendProgress
        - `operation_id` string, uuid, required: UUIDv7, lowercase canonical form.
        - `state` 'queued' | 'capacity_pending' | 'running' | 'succeeded' | 'failed' | 'canceled', required
        - `workspace_state` 'creating' | 'starting' | 'running' | 'suspending' | 'suspended' | 'resuming' | 'forking' | 'stopping' | 'failed' | 'deleting' | 'deleted', required
        - `phase` string, required: Actual snapshot preparation/capture phase, or an explicitly unavailable control status.
        - `cancelable` boolean, required: True only while the host can acknowledge cancellation before its capture cut.
        - `supported` boolean, required
        - `safe_action` 'wait_for_terminal' | 'resume' | 'retry_tools' | 'inspect_failure', required
        - `phase_started_at` string, date-time: RFC 3339 UTC timestamp with Z.
        - `deadline_at` string, date-time: RFC 3339 UTC timestamp with Z.
        - `accepted` boolean
    - `pending_reason` string, nullable, required: Why the active operation is waiting (capacity_pending reason), else null.
    - `forked_from_workspace_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
    - `disk_layout` 'legacy' | 'layered', required: legacy: one disk (template copy + changes). layered: the template chain read-only plus a workspace layer holding only the changes. Immutable.
    - `lifetime` 'persistent' | 'session', required: persistent: kept until deleted (default). session: discarded when the session ends (close(), idle timeout or draft discard;).
    - `purpose` 'standard' | 'template_draft' | 'template_test', required: standard, template_draft (a template’s dev-mode draft) or template_test (a test instance of a draft state;).
    - `idle_timeout_seconds` integer, nullable, required: Sessions only: the session ends after this long without activity (template default, else 600). Null for persistent workspaces.
    - `idle` object, nullable, required: Automatic suspend of a persistent workspace. Null for session workspaces.
      - `policy` string, required: The effective policy: adaptive, never or fixed:<seconds>.
      - `source` 'workspace' | 'template' | 'default', required: Where the policy comes from.
      - `mode` 'adaptive' | 'fixed' | 'never', required
      - `timeout_seconds` integer, nullable, required: The timeout applied to the current idle period while the workspace runs (as of the idle loop’s last decision), else null.
      - `basis` string, nullable, required: What the timeout comes from, e.g. "learned from 37 idle periods (active-hours)", "template prior", "default".
      - `next_eligible_at` string, date-time, nullable, required: When the workspace becomes eligible for automatic suspend if nothing happens before (work signals always win).
      - `suspend_request` SuspendRequest, required: A pending suspend-when-idle request: once the workspace has been idle for after_seconds, it is suspended. A tool call after requested_at (the next turn) or a resume cancels it; other work only defers it.
        - `requested_at` string, date-time, required: When the request was made; a newer request replaces it.
        - `after_seconds` integer, required
        - `not_before` string, date-time, required: requested_at + after_seconds: the earliest suspend (within a few seconds of activity flush grace after it). A running command, an attached stream or a keepalive still active then defers the suspend until after_seconds after it ends.
    - `origin` union, required: Where the workspace’s disk came from; null for a workspace opened from its template.
      - object: Forked from workspace_id.
        - `kind` 'fork', required
        - `workspace_id` string, uuid, required: UUIDv7, lowercase canonical form.
      - object: A test instance of a draft state: the draft and the captured state (checkpoint_id is null until the cell placed an instance whose state is being captured).
        - `kind` 'draft_state', required
        - `template_id` string, uuid, required: UUIDv7, lowercase canonical form.
        - `draft_workspace_id` string, uuid, required: UUIDv7, lowercase canonical form.
        - `checkpoint_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
        - `captured_at` string, date-time, nullable, required: RFC 3339 UTC timestamp with Z.
      - object: A test instance of a template version, published or not: a fresh workspace on that version.
        - `kind` 'template_version', required
        - `template_id` string, uuid, required: UUIDv7, lowercase canonical form.
        - `version` integer, required
    - `ended_reason` 'closed' | 'idle_timeout' | 'draft_discarded', nullable, required: How a session ended (closed, idle_timeout, draft_discarded); null otherwise.
    - `dev_template_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
    - `startup` WorkspaceStartup, required: Start commands and services of the workspace’s template version; null when the version has none.
      - `state` 'pending' | 'running' | 'ready' | 'failed', required: pending: a start is in progress and its startup has not begun (or never ran); running: start commands, services or readiness waits are running; ready: every step succeeded; failed: a step failed (the workspace still runs so it can be inspected; the next open runs the failed step again).
      - `trigger` 'create' | 'boot' | 'resume', nullable, required: create: a new workspace layer; boot: a cold boot; resume: a memory restore. null while pending.
      - `step` string, nullable, required: The start command (or `services`, `ready:<service>`) running or failed.
      - `service` string, nullable, required: The service that failed or is awaited.
      - `exit_code` integer, nullable, required
      - `output_tail` string, nullable, required: Up to the last 4096 bytes of the failed step’s output (stdout and stderr interleaved).
      - `reason` string, nullable, required: Failure reason: startup_failed, service_not_ready, secrets_unavailable, secret_not_available or guest_feature_unavailable.
      - `operation_id` string, uuid, nullable, required: UUIDv7, lowercase canonical form.
      - `at` string, date-time, nullable, required: RFC 3339 UTC timestamp with Z.
    - `mode` 'processful' | 'file_first', required: processful (default): one VM keeps processes, memory and files between calls; it can be suspended, resumed and forked. file_first: the state is a versioned file tree (tree_revision); there is no VM between executions, each exec runs in a fresh VM and publishes the changed files as the next revision, nothing but files survives an execution. A file-first workspace is ready (running) from creation and is never suspended; suspend, resume, snapshot, fork, reset, save-as-template, volumes and idle policies are 409 not_supported_for_mode. Immutable.
    - `tree_revision` integer, required: file_first: the latest revision of the file tree (0 = the empty tree the workspace starts with; each mutating file call or execution publishes the next one; cell file responses carry it as X-Tree-Revision). Always 0 for processful workspaces.
    - `upgrade_available` object, nullable: Supported target dimensions of an opt-in upgrade; null dimension stays unchanged.
      - `memory_layout` integer, nullable, required
      - `guest_revision` integer, nullable, required
      - `template_version` integer, nullable, required
    - `upgrade_pending` object, nullable: The pinned upgrade target awaiting its cold start.
    - `computer_use` ComputerUse, required
      - `enabled` boolean, required: Effective: tool tokens carry the computer tool (the workspace’s value, else the template’s, and the version runs a desktop).
      - `workspace` boolean, nullable, required: The workspace’s own value; null follows the template.
      - `template` boolean, required: The template’s switch for its workspaces.
      - `available` boolean, required: The workspace’s template version can run a desktop (its guest agent has desktop.v1).
    - `created_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `updated_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `deleted_at` string, date-time, nullable, required: RFC 3339 UTC timestamp with Z.
  - `source_workspace_id` string, uuid, required: UUIDv7, lowercase canonical form.
  - `cell_endpoint` string, nullable, required
  - `tool_token` ToolToken, required
    - `token` string, required: Compact ES256 JWS; send as `Authorization: Bearer` to the cell gateway.
    - `token_type` 'Bearer', required
    - `expires_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `issued_at` string, date-time, required: RFC 3339 UTC timestamp with Z.
    - `kid` string, required
    - `audience` string, required
    - `workspace_id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `agent_session_id` string, uuid, required: UUIDv7, lowercase canonical form.
    - `ownership_epoch` integer, required
    - `tools` string[], required
    - `cell_endpoint` string, required

## Other responses

- `202`: Default Response
- `4XX`: Default Response
- `5XX`: Default Response

---

[API](https://skmtc.dev/shardflux/apis/shardflux-api.md) · [All operations](https://skmtc.dev/shardflux/apis/shardflux-api/llms.txt) · [OpenAPI document](https://skmtc.dev/shardflux/apis/shardflux-api/revisions/ea685f7b39c4?raw)