CLI reference (main.py)

All flags below are defined in agentic-orchestration-tool/main.py (parse_args). Paths are relative to the tool root unless absolute.

Positional

Argument Meaning
TASK Optional. If present without --dynamic*: router task string. If with --dynamic*: user goal. If omitted: interactive router loop (or batch with --batch).

Modes

Flag Meaning
--batch Run once and exit when TASK omitted (uses --config).
--interactive (-i) Interactive loop on a fixed --config workflow.
--dynamic Plan + run ephemeral multi-step workflow from TASK.
--dynamic-iterative Stepwise dynamic runs with replanning.
--example healthcare|logistics Apply built-in vertical overlay (context + extra catalogs) without manual .env path wiring.

Workflow / router

Flag Default Meaning
--config config/workflows/workflow.yaml Static workflow YAML.
--config-dir config Root for workflows/*.yaml scan.
--router-model ROUTER_OLLAMA_MODEL or llama3.2 Ollama model for routing.
--router-host OLLAMA_HOST Ollama base URL.

Output / artifacts

Flag Meaning
--output-dir DIR Save extracted files here without prompting.
--no-save Do not save extracted files.
--prompt-save Prompt for save location.
--no-verify Skip post-save npm verify (AGENTIC_VERIFY also applies).
--quiet Less console noise; with --dynamic, skip plan/step progress on stderr.

Worker (distributed backends)

Flag Meaning
--execute-step SPEC.JSON Worker mode: run one step from a StepSpec JSON file and exit. Used by subprocess/K8s backends; not for normal CLI runs.

Dynamic catalog paths

Flag Default Meaning
--agent-providers-catalog (--providers-catalog) config/agent_providers Agent YAML dir or bundle.
--mcp-providers-catalog config/mcp_providers MCP YAML dir or bundle.
--dynamic-agent-provider-ids empty Restrict planner choices to comma-separated provider IDs.
--dynamic-attachments none Attachment manifest for file-aware dynamic planning/execution.

Iterative dynamic

Flag Meaning
--dynamic-iterative-rounds N Max rounds before synthesis.
--dynamic-iterative-auto Enable controller between rounds.
--dynamic-iterative-max-rounds N Cap with auto.
--dynamic-iterative-min-rounds N Minimum rounds before stop allowed.
--dynamic-iterative-no-synthesize Skip final synthesis.

Session

Flag Meaning
--orchestrator-session NAME Session slug for __orchestrator_sessions__/.
--orchestrator-session-reset Delete session JSON before run.

Platform agent harness

Flag Meaning
--harness-agent ID Run harness for one catalog agent_provider_id and exit.
--harness-batch Run harness for all (or filtered) catalog agents and exit.
--harness-tier TIER static (L0), connectivity (L1), smoke (L2), capability (L3). Aliases: l0–l3. Default: env AGENTIC_HARNESS_TIER or static.
--harness-filter GLOB With --harness-batch: fnmatch on provider ids (e.g. gpt_*).
--harness-max-agents N Cap batch size.
--harness-profile PROFILE Force profile (general, research, write, reason, coding, vision).
--harness-backend NAME inprocess (default) or subprocess for L2/L3.
--harness-json Emit JSON report on stdout.
--harness-fail-fast Stop batch on first failure.
--harness-verbose Verbose CrewAI output during smoke runs.

Env: AGENTIC_HARNESS_TIER, AGENTIC_HARNESS_EVAL, AGENTIC_HARNESS_EVAL_MODEL, AGENTIC_HARNESS_SKIP_SELFCONTAINED_INIT, AGENTIC_HARNESS_FEED_PLANNER, AGENTIC_HARNESS_RECORD_STATS. See Agent harness roadmap.