Generate, Run & Replay

The core generate → run loop, plus exporting chains to disk and replaying past runs.

care generate "<task>"

Generate a CARL chain from a free-form task via MAGE.

care generate "weather report for SF" --save weather --output weather.json

Flag	Purpose
--mode fast|deep	MAGE generation mode (defaults to MAGE’s own default).
--save NAME	Save the generated chain to Memory under NAME (new entity).
--output PATH	Write the chain to a file (format inferred from .json / .py).
--output-format json|python	Override the export format.
--json	Emit the chain as JSON instead of a summary line.

If a generate call fails because of authentication or an expired token, the CLI now prints a single friendly hint pointing you at care init (to re-seed your credentials) and care doctor (to diagnose the endpoint) — instead of a raw traceback.

Looking for care export?

For a one-off file, export the generated chain inline with generate --output PATH or a saved chain with run --export PATH (below) — both write .json or .py (a runnable Python module) based on the extension. To ship several chains (plus their AgentSkills) as one portable tarball, use the dedicated care export command.

care run <chain_id>

Fetch a saved chain from Memory, preflight it, and optionally execute it.

care run my-chain --execute --input city=Paris --save-result paris-run
care run my-chain --export chain.py            # export, don't run

Flag	Purpose
--channel NAME	Version channel to read (default latest).
--execute	Execute via CARL after preflight (needs mmar_carl + an LLM key).
--task TEXT	Override outer_context for the run.
--input KEY=VALUE	Add a pair to context.memory['input'] (repeatable).
--save-result NAME	Persist a run digest as a memory_card (requires --execute).
--export PATH	Also write the fetched chain to a file (.json / .py).
--export-format json|python	Override the export format.
--json	Emit the preflight result as JSON.
--log PATH / --log-level debug|info	Write a structured debug log of the run.

care replay <source>

Step through a saved ReasoningResult / RunRecord JSON from a previous run. Use - to read from stdin.

care replay run.json            # whole session, one block per step
care replay run.json --step 2   # just step 2 (0-indexed)
cat run.json | care replay -

Flag	Purpose
--step N	Render only step N (0-indexed); default renders the whole session.
--json	Structured payload (chain metadata + every step) instead of the walkthrough.

Run care <command> --help for the authoritative, up-to-date flag set.