Skip to main content
Agents should call: 
sf describe --all --json
sf tools plan "first time setup and login" --json
sf tools search "portfolio history" --json
sf tools plan "explain why CPI residual moved" --json
sf tools plan "tell me if oil explains CPI residual and send it to telegram" --json
sf tools plan "get my current user state and find quality opportunities" --json
sf guide --agent --json
sf describe --all --json returns the machine-readable catalog of CLI tools. sf tools search ranks that catalog for a natural-language task. sf tools plan returns a conservative ordered sequence with command strings, auth requirements, side-effect levels, and reasons. sf guide --agent --json returns local playbooks built around the same command contract. This page documents CLI command discovery only. It is not the SDK/Agent contract manifest. Use these surfaces for the different tool universes:
SurfacePurpose
GET /api/contracts/toolsStrict SDK/Agent contract manifest with canonical dotted names
sf describe --all --jsonLocal CLI command manifest
GET /api/toolsBroad hosted compatibility inventory

​
Manifest fields

type CliToolManifestItem = {
  name: string
  namespace?: string
  command: string
  description: string
  authRequired: boolean
  authType: 'none' | 'sf_api_key' | 'exchange_key' | 'browser_session' | 'mixed'
  readOnly: boolean
  sideEffects: boolean
  longRunning: boolean
  jsonStable: boolean
  outputModes: ('human' | 'json' | 'markdown')[]
  dangerLevel: 'safe_read' | 'account_read' | 'local_runtime' | 'server_write' | 'exchange_write'
  requiredArgs: Array<{ name: string; description: string }>
  options: Array<{ name: string; type: string; default?: unknown; description: string }>
  examples: Array<{ command: string; purpose: string }>
  responseSchema?: unknown
  errorSchema?: unknown
}

​
Safety model

Danger levelMeaning
safe_readPublic or local read-only data
account_readReads authenticated user data
local_runtimeStarts or stops a local process
server_writeMutates SimpleFunctions server state
exchange_writeCan place/cancel exchange orders
Agents should prefer safe_read and account_read commands unless explicitly authorized to mutate state.

​
Planning model

Agents should use the catalog before guessing commands. 
sf tools plan "watch this FRED release and alert me when related markets move" --json
sf tools plan "find quality opportunities for my agent loop" --json
sf tools plan "tell me if oil explains CPI residual and send it to telegram" --json
sf tools plan "tell me if oil explains CPI residual and send it to telegram" --allow-side-effects --json
sf workflow demo monitor --dry-run --json
The planner is local and manifest-backed. It should surface read-first steps such as status, login, setup --check, doctor, guide, brief, discover, investigate, query, econ, policy, world, inspect, watchlist, alerts, webhooks, and monitor list; write/runtime/trade commands remain marked with side-effect metadata so the caller can require explicit approval. Discovery tasks route through sf discover --quality --json, which combines ideas, world opportunities, cross-venue gaps, fresh listings, and contagion signals into a ranked queue. Authenticated loop tasks route through sf brief --agent --json, which summarizes user/workflow/portfolio/world context without returning secrets or raw delivery payloads. Onboarding tasks route through sf status --json, sf login, sf setup --check, sf doctor --agent --json, sf guide --agent --json, sf brief --agent --json, and sf describe --all --json. Monitoring and delivery tasks such as “tell me if X and send it to Telegram” route through read-first investigation and existing workflow reads, then gate sf monitor create and sf monitor run behind --allow-side-effects. The workflow demo is also local and non-mutating. It emits the full watchlist, alert, webhook, and research-monitor chain as a dry run so an agent can show the user exactly what would happen before running authenticated writes.

​
Workflow namespaces

The live manifest includes the workflow namespaces used by monitoring agents:
NamespaceTypical commandDanger model
watchlistsf watchlist list --jsonread plus authenticated server writes for add/remove/refresh
alertssf alerts list --jsonread plus authenticated server writes for create/pause/test/delete
webhookssf webhooks list --jsonread plus authenticated server writes for add/test/rotate/delete
monitorsf monitor list --jsonread plus authenticated server writes for create/run/pause/resume/delete
polysf poly search "query" --jsonpublic/wallet-address Polymarket global reads
polyussf polyus markets --jsonpublic Polymarket US reads; auth boundary is explicit
forumsf forum read --jsonauthenticated reads and channel/message writes
briefsf brief --agent --jsonauthenticated read-only agent loop brief
discoversf discover --quality --jsonpublic read-only quality discovery feed
guidesf guide --agent --jsonlocal read-only onboarding playbooks
workflowsf workflow demo monitor --dry-run --jsonlocal read-only dry-run plan
tracesf trace receipt trace.ndjson --jsonlocal read-only trace audit