Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/shobcoder/shob/llms.txt

Use this file to discover all available pages before exploring further.

shob run is the primary way to interact with the Shob AI agent from the command line. It accepts a free-text message, dispatches it to the configured AI model, and streams the agent’s output back to your terminal. You can continue existing sessions, attach to a remote Shob server, attach files to your message, and control output formatting — all from a single command.

Usage

shob run [message..]
The message argument is a positional, variadic string. You can pass multiple words directly as arguments; Shob joins them into a single message.
shob run "explain the purpose of every function in src/auth.ts"
shob run explain the purpose of every function in src/auth.ts

Flags

--command
string
A named command to run; the message positional is passed as arguments to that command.
--continue
boolean
Alias: -cContinue the most recent session rather than creating a new one. Useful for iterating on a task without repeating context.
--session
string
Alias: -sContinue a specific session identified by its session ID. Use shob session list to find session IDs.
--fork
boolean
Fork the session before continuing. Requires --continue or --session. Creates a new branch of the conversation rather than appending to it.
--share
boolean
Share the session by generating a shareable link after it completes.
--model
string
Alias: -mThe model to use, in provider/model format — for example anthropic/claude-opus-4-5 or openai/gpt-4o. Overrides any model set in the project configuration.
--agent
string
The name of a custom agent to use for this run. Custom agents are created with shob agent create.
--format
string
default:"default"
Output format. Accepted values:
  • default — human-readable, formatted output streamed to the terminal
  • json — raw JSON events, suitable for piping to other tools
--file
string[]
Alias: -fOne or more file paths to attach to the message. Can be specified multiple times: --file a.ts --file b.ts.
--title
string
An explicit title for the session. If omitted, Shob uses a truncated version of the message as the title.
--attach
string
Attach to a running Shob server instead of starting a local agent. Pass the base URL of the server, for example http://localhost:4096. When this flag is set, the agent runs on the remote server.
--password
string
Alias: -pBasic auth password for the remote Shob server. Falls back to the SHOB_SERVER_PASSWORD environment variable when not provided.
--dir
string
Working directory for the agent. When used with --attach, this is interpreted as a path on the remote server.
--port
number
Port for the local embedded server. Defaults to a random available port when not specified.
--variant
string
Model variant for provider-specific reasoning effort settings, for example high, max, or minimal. The available values depend on the chosen provider and model.
--thinking
boolean
default:"false"
Show thinking blocks in the output when the model supports them (e.g. extended thinking on Claude models).
--dangerously-skip-permissions
boolean
default:"false"
Auto-approve all permission prompts that are not explicitly denied. Use with extreme caution — this allows the agent to take destructive actions without confirmation.

Examples

Basic task
shob run "refactor the authentication module"
Continue the last session
shob run -c "also add tests for the new functions"
Use a specific model
shob run -m anthropic/claude-opus-4-5 "explain what this codebase does"
Attach to a remote Shob server
shob run --attach http://localhost:4096 "fix the failing tests"
JSON output for scripting
shob run --format json "list all exported functions" | jq '.content'
Attach files to the message
shob run -f src/auth.ts -f src/auth.test.ts "improve test coverage"
Fork the last session and try a different approach
shob run -c --fork "try a different implementation strategy"
Using --dangerously-skip-permissions disables all interactive permission checks. Only use this in controlled environments (e.g. CI) where you fully trust the agent and have reviewed the task scope.

Build docs developers (and LLMs) love