Read custom operator source by node identity

GET /workflow/v2/custom-operators/source requires exactly one nonempty X-Workspace-ID. Supply canonical node_id and version, with optional task_id. The result envelope uses string code: "OK", msg: "OK", and data.

Two sources

With nonempty task_id, only the task workspace source is selected, returning origin: "workspace". The task must belong to the current workspace and have context. Task and context are rechecked before reading. The loader currently accepts only version=draft, with node prefix moi:custom.operator:{workspace_id}: followed by a valid identifier. It reads operator.json and main.py from that task workspace’s operator directory. Input/output schemas must be objects and the manifest is strictly validated. Identity, handler, and language derive from the trusted workspace, currently Python and main.handle. Missing task files do not fall back to saved operators.

On first workspace read, this GET may restore a local directory from a persisted snapshot and write a readiness manifest. It is therefore not entirely free of local filesystem writes. It neither publishes the operator nor starts execution.

Without task_id, exact node_id + version lookup finds a persisted operator in the current tenant and returns origin: "published". Up to two rows are read to detect ambiguity; exactly one is required. published is a protocol source value, not independent proof that the operator is enabled or the workflow published. The operator must be a code type with nonempty SourceFileID. Catalog supplies the source, and whitespace-only content is invalid. Current detail reads do not separately filter disabled state; do not invent an enabled-only restriction.

Each source passes its generated authorization boundary. Internal Platform calls selecting a source do not make the HTTP endpoint authorization-free.

Request example

Obtain exact node identity and version from an existing workflow node. Add the current authoring task ID only when reading a task workspace draft:

curl --get "$PRODUCT_API_BASE_URL/workflow/v2/custom-operators/source" \
  -H "X-API-Key: $PRODUCT_API_KEY" \
  -H "X-Workspace-ID: $WORKSPACE_ID" \
  --data-urlencode "node_id=$NODE_ID" \
  --data-urlencode "version=$NODE_VERSION" \
  --data-urlencode "task_id=$AUTHORING_TASK_ID"

Omit the last parameter when reading a persisted operator. task_id selects the source; it is not a fallback hint after a draft read fails.

PRODUCT_API_BASE_URL uses the product API prefix https://api.moi.matrixorigin.cn/v5.

Field

Type

Required

Constraints

node_id

string

Yes

Canonical node identity, not an integer operator ID. Exactly once, nonempty, without surrounding whitespace

version

string

Yes

Exact version for that identity. Exactly once, nonempty, without surrounding whitespace

task_id

string

No

At most once; nonempty values cannot have surrounding whitespace. Empty means omitted

Response fields

Field

Type

Description

data.source

string

Complete source text, potentially containing sensitive business information. Store according to code and secret handling rules

data.origin

string

workspace or published

data.node_id

string

Node identity

data.version

string

Node version

data.language

string

Optional source language

data.handler

string

Optional entry function

data.name

string

Optional operator name

data.identifier

string

Optional operator identifier

Errors and pending verification

Duplicate/empty parameters, noncanonical whitespace, and missing source map to 400/ErrParamInvalid; missing workspace headers map to 400/ErrWorkspaceRequired. Missing persisted identities/tasks and task mismatches follow service error mappings. Ambiguous identities and wrapped source-read errors map to 500/ErrServer. Raw file-read errors in task directories are not consistently converted to ErrNotFound; do not describe every missing-file case as404.

Frontend helper comments still describe draft-first with published fallback, unlike current service branches. Existing Go/Python/CLI fixture tests still use origin=draft. Both are maintenance candidates; passing tests do not establish protocol enums.

Last updated on