Skip to main content

​
Why agents like defi-cli

  • Stable envelope schema (version, success, error, meta)
  • Stable exit code mapping (including execution-specific codes 20-24)
  • Canonical chain and asset IDs
  • Provider-scoped retrieval IDs on lend/yield rows (provider_native_id, provider_native_id_kind)
  • Deterministic --select projections
  • Machine-readable command schema (defi schema)
  • Execution commands (plan, run, submit, status) for on-chain actions
  1. Resolve all chain/asset inputs (assets resolve where useful).
  2. Execute command with --json --results-only for success payloads.
  3. Handle non-zero exit codes using error.code and error.type.
  4. Persist canonical IDs and base-unit amounts.
For execution workflows:
  1. Use plan to dry-run and inspect steps before committing.
  2. Use actions list / actions show / actions estimate to inspect persisted actions.
  3. Use submit --action-id to broadcast a planned action (requires signer).
  4. Use --input-json / --input-file for structured input on plan and submit.

​
Suggested defaults

--timeout 12s --retries 2 --max-stale 5m
For strict workflows: 
--strict --no-stale

​
Command allowlist mode

Use --enable-commands to constrain executable commands in agent sandboxes. 
defi --enable-commands "providers list,yield opportunities,swap plan" providers list --results-only
If a command is blocked, exit code 16 (command_blocked) is returned. This is especially useful for restricting which execution commands an agent can call.