Node reference

Choose a node by what it does, then make its inputs, routes, and side effects explicit.

Each reference page contains a complete source example, the node's configuration fields and defaults, its input ports and output fields, and its preflight requirements. Tables follow the installed Rattish machine contracts.

Commands and agents

Node	Purpose
bash-command	Run one command and capture its result
shell-script	Run a maintained Bash script
python-script	Run a maintained Python script
agent	Ask a provider to perform an authored task
common-llm-task	Use a review, summary, extraction, or other common prompt pattern

Files and resources

Node	Purpose
read-file	Read text and expose its metadata
write-file	Create, replace, or append text
copy-file	Copy a file, folder, or symlink
move-file	Move or rename a resource
delete-file	Remove a resource with trash and recursion policies
file	Describe a file without reading its content
folder	Describe a directory without enumerating it
open-resource	Open a file, folder, URL, or application
prompt-file	Render a prompt template to a file

Control and interaction

Node	Purpose
loop	Fan out runtime work items
break	Stop a loop's pending iterations
workflow	Call a child workflow's public interface
approval-gate	Obtain a human decision
notification	Notify a desktop or external channel
rem-prompt	Ask Rem and use its reply
rem-task	Run one of Rem's tasks and wait for its result
http-request	Call a service and expose its response

Local retrieval

Node	Purpose
local-vectorize	Build or validate an on-disk text index
local-search	Retrieve ranked chunks from that index

These nodes are distinct from Rem's conversational tools. Enabling Second Brain in a thread does not add a node to a workflow, and declaring an HTTP node does not enable Rem web search.

Shared node fields

Every node declares type. The following execution and routing fields are shared:

Field	Default	Meaning
needs	[]	One-time readiness dependencies
to	No routes	Explicit outgoing routes, including conditional ones
with	{}	Input-port and local bindings
start	false	Start assertion/presentation hint; cannot have needs
finish	None	pass or fail endpoint; cannot have outgoing routes
allow-fail	false	Allow an operation failure to participate in recovery routing
timeout	None	Authored duration limit
max-runs	None	Maximum runs of the node
max-concurrency	1	Maximum overlapping activations of the node
retry-count	0	Additional operation retries
retry-delay	1s	Delay between retries

finish: fail cannot also declare allow-fail. Node contracts can add restrictions: for example, a break node cannot have outgoing routes. A loop's source concurrency is separate from the shared activation limit.

Source spelling and selectors

Rattish configuration uses hyphens, such as working-dir and create-dirs. Output selectors use the schema's exact names, often underscores, such as node.check.output.exit_code.

Do not invent node types named start, pass, fail, branch, or join. Use shared fields and routes/readiness for those behaviors.

Check an example

Copy a complete example into workflow.rattish in a disposable project and create any prerequisite files mentioned on its page. Check and preflight do not execute nodes:

Raticode rattish format workflow.rattish
Raticode rattish check workflow.rattish --format json
Raticode rattish preflight workflow.rattish --format json

An environment-dependent preflight diagnostic can mean a provider, notifier, interpreter, index, or opener is missing. Resolve that requirement before execution rather than bypassing the check.

See Validation and preflight and Inputs and bindings.