@slatehq/agent-cli

Agent-first CLI for discovering blocks, compiling wf-ir@1, applying workflows, and running/inspecting executions.

Core commands

slate-cli auth login --token <slat_...> [--profile default] [--endpoint https://api.slatehq.ai]
slate-cli status [--profile default] [--format json]

slate-cli wf blocks [--ids googleSearch,webScrape,llm] [--runtime all|supported|unsupported] [--format json|markdown]
slate-cli wf validate --ir /abs/path/workflow.wf-ir.json --format json
slate-cli wf compile --ir /abs/path/workflow.wf-ir.json --out /tmp/workflow.graph.json --format json
slate-cli wf apply --ir /abs/path/workflow.wf-ir.json --mode create-new --publish --format json
slate-cli wf run --workflow-id <id> --inputs-json '{"query":"best ai seo tools"}' --check-inputs true --wait true --format json
slate-cli wf inspect --run-id <run-id> --format json

Lifecycle

1. wf blocks Returns full discoverable catalog (purpose-first rows).

2. wf blocks --ids ... Use detail mode to load schema/constraints only for selected blocks.

3. Author wf-ir@1 IR is the canonical authoring format for agents.

4. wf validate Lints IR and returns required runtime inputs, issues, fingerprint, and normalized IR. Add --strict only for optional hardening gates.

5. wf compile Deterministically compiles IR into backend graph JSON.

6. wf apply Creates/patches workflow and optionally publishes.

7. wf run + wf inspect Runs and returns terminal status; inspect loads failure step context.

Notes

• wf-ir@1 supports step, forEach, and if flow nodes.
• wf export --to ir is best-effort linear export from graph JSON.
• Manifest drift check:
  • npm run check:manifest