agnify pipeline

agnify pipeline get

Print the target execution as JSON.

Defaults to $AGNIFY_DRAFT_EXECUTION_ID (the session’s origin draft); pass --execution-id to read a different draft (e.g. a clone the improvement loop just created).

Usage

agnify pipeline get [OPTIONS]

Options

Flag	Description
--execution-id, -e	Execution to target. Defaults to $AGNIFY_DRAFT_EXECUTION_ID (the session’s origin draft). Pass this to edit a freshly cloned draft in the improvement loop, since the subprocess env is fixed at spawn.

agnify pipeline pull

Print the current draft execution as JSON. Alias for get.

Usage

agnify pipeline pull [OPTIONS]

Options

Flag	Description
--execution-id, -e	Execution to target. Defaults to $AGNIFY_DRAFT_EXECUTION_ID (the session’s origin draft). Pass this to edit a freshly cloned draft in the improvement loop, since the subprocess env is fixed at spawn.

agnify pipeline push

Replace the draft’s whole node snapshot with SNAPSHOT. Alias for set.

Usage

agnify pipeline push [OPTIONS] SNAPSHOT

Arguments

Name	Description
SNAPSHOT	A snapshot as inline JSON (starts with ’{’ or ’[’) or the path to a JSON file. Either a bare list of nodes or an object with a ‘nodes’ field is accepted.

Options

Flag	Description
--execution-id, -e	Execution to target. Defaults to $AGNIFY_DRAFT_EXECUTION_ID (the session’s origin draft). Pass this to edit a freshly cloned draft in the improvement loop, since the subprocess env is fixed at spawn.

agnify pipeline set

PUT /executions/{target}/config with the snapshot.

Targets $AGNIFY_DRAFT_EXECUTION_ID by default; pass --execution-id to edit a clone in the improvement loop. Only DRAFT executions are reconfigurable, so target a fresh clone, not a completed run.

The snapshot replaces the draft’s node list wholesale, so it must carry every node you want the run to keep, not just the ones you changed. agnify pipeline pull prints the whole execution, so extract its pipeline_snapshot field (jq .pipeline_snapshot) and edit that — piping pull straight back in is rejected as the wrong shape.

The local validator runs first and short-circuits on any issues — the backend would catch most of these at the Pydantic boundary, but the local check keeps the round-trip cost on the caller’s side and produces a list of problems instead of the first one.

Usage

agnify pipeline set [OPTIONS] SNAPSHOT

Arguments

Name	Description
SNAPSHOT	A snapshot as inline JSON (starts with ’{’ or ’[’) or the path to a JSON file. Either a bare list of nodes or an object with a ‘nodes’ field is accepted.

Options

Flag	Description
--execution-id, -e	Execution to target. Defaults to $AGNIFY_DRAFT_EXECUTION_ID (the session’s origin draft). Pass this to edit a freshly cloned draft in the improvement loop, since the subprocess env is fixed at spawn.

agnify pipeline validate

Validate a snapshot locally — no HTTP, no DB.

On success, prints {"ok": true, "node_count": N}. On failure, prints {"ok": false, "issues": [...]} to stderr and exits non-zero so a caller can iterate without a round-trip.

Usage

agnify pipeline validate SNAPSHOT

Arguments

Name	Description
SNAPSHOT	A snapshot as inline JSON (starts with ’{’ or ’[’) or the path to a JSON file. Either a bare list of nodes or an object with a ‘nodes’ field is accepted.