Pseudo CLI Command Format

Pseudo CLI commands use a registry format rather than shell parsing.

Base Syntax

codebolt <module> <action> [--param value ...] [--flag] [--no-flag]

The parser splits the command into:

• module: the top-level registry entry such as fs, git, terminal, or mcp
• action: the command within that module such as read, status, exec, or execute
• params: named flags converted into a key-value object

Supported Parameter Forms

Form	Example	Parsed value
String flag	--path "./src/index.ts"	path: "./src/index.ts"
Boolean true	--recursive	recursive: true
Boolean false	--no-recursive	recursive: false
Number	--max 5	max: 5
JSON object	--data {"status":"done"}	data: { status: "done" }
JSON array	--paths ["a.ts","b.ts"]	paths: ["a.ts", "b.ts"]

Parsing Rules

• Single and double quoted strings are supported.
• Backslash escaping inside quoted strings is supported.
• JSON blobs starting with { or [ are tokenized as a single value and then JSON.parse() is attempted.
• Literal true and false are converted to booleans.
• Numeric tokens matching -?\d+(\.\d+)? are converted to numbers.
• Positional arguments after <action> are ignored. Use named flags instead.

Examples

Filesystem

codebolt fs read --path "./package.json"
codebolt fs list --path "./src" --recursive
codebolt fs read-many --paths ["package.json","tsconfig.json"]

Git

codebolt git status
codebolt git commit --message "docs: add pseudo cli reference"
codebolt git clone --url "https://github.com/example/repo.git" --path "./repo"

Terminal

codebolt terminal exec --command "npm run build"

Pseudo CLI itself does not interpret shell syntax like pipes or redirects. If you need actual shell behavior, pass the full shell command as the --command value to terminal exec.

Browser

codebolt browser navigate --url "https://example.com"
codebolt browser click --element "Login"
codebolt browser scroll --direction "down" --pixels 800

Task And Thread Operations

codebolt task create --title "Write docs" --description "Document pseudo cli"
codebolt thread create-start --task "Review the generated docs" --agent "agent-id"
codebolt task update --id "task-123" --data {"status":"completed"}

Help Behavior

The pseudo CLI source includes help generation for these patterns:

codebolt help
codebolt help <module>
codebolt <module>

Intended behavior:

• codebolt help lists all registered modules
• codebolt help fs shows commands available for the fs module
• codebolt fs shows module help when no action is provided

Validation And Errors

Before execution, the registry checks required flags from each command definition.

Common failure cases:

• Unknown module: the module is not in commandRegistry
• Unknown action: the module exists but the action does not
• Missing required parameter: a required --flag is absent
• Execution failure: the underlying SDK module throws or returns an error-shaped response

Formatter behavior:

• Successful results are returned as llmContent plus a shortened returnDisplay
• Missing params and unknown commands are marked as INVALID_TOOL_PARAMS
• Execution failures are marked as EXECUTION_FAILED

Interception Note

The terminal tool only bypasses real shell execution when isCodeboltCommand() matches the command string. In the current source, that matcher checks for codeboltcmd , while the parser and help text use codebolt .... Keep that distinction in mind when testing end-to-end interception.