Configuration reference

This is the lookup layer for the Strands harness factory (create_harness in Python, createHarness in TypeScript). Every argument is optional. The two libraries expose the same surface, each spelled for its language (snake_case in Python, camelCase in TypeScript). For how each option is used in practice, follow the links into the configure and tools guides.

Options

Python	TypeScript	Default	Purpose
model	model	bedrock/global.anthropic.claude-opus-4-8	A provider/name string, a bare Amazon Bedrock id, or a Model instance. See choose a model.
effort	effort	"auto"	Reasoning effort: "auto", "low", "medium", "high", or off. Ignored for a Model instance.
instructions	instructions	none	A domain block appended after the harness contract. Ignored when a full system prompt is passed.
tools	tools	none	Consumer tools, added alongside the built-ins. See add tools.
plugins	plugins	none	Consumer Strands Harness SDK plugins, added alongside the built-in plugins.
mcp_servers	mcpServers	none	MCP servers to connect: a JSON file path or a mapping. See MCP servers.
builtin_tools	builtinTools	shell, read, write, edit, web_fetch, web_search, programmatic_tool_caller, subagent	Built-in tools to enable, or [] for none.
builtin_tools={"web_fetch": {"model": ...}}	builtinTools: { web_fetch: { model } }	provider default	The summarizer model web_fetch runs on. See web access.
caching	caching	"auto" (on)	Prompt caching where the provider supports it; off disables what Strands harness configures.
context_manager	contextManager	"auto"	"auto", "agentic", or off. Enables context management and offloading.
session={"id": ...}	session: { id }	none	Persist and resume this conversation by id.
session={"dir": ...}	session: { dir }	./.agent/sessions	Where session state and offloaded artifacts live.
skills	skills	./.agent/skills	Directory (or list) scanned for Agent Skills; off to disable.
builtin_plugins	builtinPlugins	["todos", "environment"]	Built-in feature plugins to enable, or [] for none.
memory	memory	on	File-based long-term memory; off to disable.
memory={"dir": ...}	memory: { dir }	./.agent/memory	Where the default memory store’s files live.
memory={"stores": [...]}	memory: { stores }	none	Swap the memory backend while keeping Strands harness’s policy.
interventions	interventions	none	Gate tool calls behind approval or a policy. See interventions.
background_tasks	backgroundTasks	{ agentic: ['*'] }	Background-tasks policy. See background tasks.

Value notes

• effort accepts "auto" (provider recommended), "low"/"medium"/"high" (explicit; some providers accept finer levels), and "off". A level the resolved provider rejects fails at construction.
• caching accepts "auto"/True (on), False/None (off). Off has no effect where caching is automatic (OpenAI, Google, bedrock-mantle). Enabling it explicitly on a Model instance is ignored with a warning.
• context_manager accepts "auto", "agentic", and False/None. Off also disables context offloading.
• memory accepts True (on) and False/None (off). Ignored when a memory_manager is passed through.
• builtin_tools, builtin_plugins each take a list of names; an unknown name fails at construction with the valid names.

Passthrough to the Agent

Any keyword the factory does not name is forwarded to the underlying Agent constructor. An explicit passthrough value takes precedence over the default it corresponds to (for example, passing system_prompt overrides the prompt Strands harness would build from instructions, and passing memory_manager overrides memory). In TypeScript, HarnessAgentOptions extends the Strands Harness SDK’s AgentConfig (minus the fields Strands harness manages), so any other AgentConfig field, such as retryStrategy, passes through verbatim. See compose with the Strands Harness SDK.

Exported building blocks

Beyond the factory, Strands harness exports the pieces it composes: HARNESS_CONTRACT, build_system_prompt / buildSystemPrompt, resolve_memory, and resolve_interventions (Python only). These are documented in compose with the Strands Harness SDK.