Runtime modes

These commands exercise the runtime session machinery. bootstrap and turn-loop build stateful sessions against the mirrored inventories. The five remote-mode commands simulate different connection modes supported by the Claude Code agent harness.

`bootstrap`

Builds a full runtime-style session report from the mirrored inventories for a given prompt.

python3 -m src.main bootstrap <prompt> [--limit N]

Arguments

prompt

string

required

The prompt to use as the session seed. The runtime routes this prompt, executes matched command and tool shims, streams a turn event sequence, and persists the session.

--limit

integer

default:"5"

Maximum number of routed matches to consider when selecting commands and tools to execute.

Output sections

bootstrap prints the RuntimeSession.as_markdown() output, which contains the following sections:

Section	Contents
Context	Python file count and archive availability flag from `build_port_context()`
Setup	Python version, implementation, platform, and test command
Startup Steps	Six-step prefetch and init sequence performed by `WorkspaceSetup.startup_steps()`
System Init	System init message generated by `build_system_init_message(trusted=True)`
Routed Matches	Each match formatted as `[kind] name (score) — source_hint`
Command Execution	Output messages from each matched command shim’s `execute()` call
Tool Execution	Output messages from each matched tool shim’s `execute()` call
Stream Events	Raw stream event types emitted by `stream_submit_message()`
Turn Result	The formatted output of the final `submit_message()` turn
Persisted session path	File path where the session JSON was saved under `.port_sessions/`
History log	Timestamped log of context, registry, routing, execution, and session store events

Example

python3 -m src.main bootstrap "read a configuration file" --limit 3

# Runtime Session

Prompt: read a configuration file

## Context
python_files=74, archive_available=False

## Setup
- Python: 3.12.2 (CPython)
- Platform: macOS-14.4-arm64-arm-64bit
- Test command: python3 -m unittest discover -s tests -v

## Startup Steps
- start top-level prefetch side effects
- build workspace context
- load mirrored command snapshot
- load mirrored tool snapshot
- prepare parity audit hooks
- apply trust-gated deferred init

## System Init
[system init message]

## Routed Matches
- [tool] Read (2) — tools/read.ts
- [command] /config (1) — commands/config.ts

## Command Execution
[/config] executed with prompt: read a configuration file

## Tool Execution
[Read] executed with payload: read a configuration file

## Stream Events
- message_start: {'type': 'message_start', ...}
- command_match: {'type': 'command_match', ...}
- tool_match: {'type': 'tool_match', ...}
- message_delta: {'type': 'message_delta', ...}
- message_stop: {'type': 'message_stop', ...}

## Turn Result
Prompt: read a configuration file
Matched commands: /config
Matched tools: Read
Permission denials: 0

Persisted session path: .port_sessions/a3f1c2d4e5b6.json

`turn-loop`

Runs a small stateful multi-turn loop against the mirrored runtime.

python3 -m src.main turn-loop <prompt> [--limit N] [--max-turns N] [--structured-output]

Arguments

prompt

string

required

The initial prompt. Turn 1 uses this prompt verbatim. Subsequent turns append [turn N] to form the turn prompt.

--limit

integer

default:"5"

Maximum number of routed matches to use when selecting commands and tools. Routing runs once before the loop begins; all turns share the same matched command and tool names.

--max-turns

integer

default:"3"

Number of turns to execute. The loop stops early if any turn’s stop_reason is not completed.

--structured-output

boolean

default:"false"

When set, each turn’s output is serialised as JSON instead of plain text.

Per-turn output format

Each turn is printed as:

## Turn N
<turn output>
stop_reason=<reason>

Stop reasons

Reason	Description
`completed`	Turn processed normally; the loop continues to the next turn
`max_turns_reached`	The engine’s internal turn counter hit `max_turns` before the prompt was processed
`max_budget_reached`	Projected token usage exceeded the engine’s `max_budget_tokens` (default 2000)

Example

python3 -m src.main turn-loop "list files in the project" --max-turns 2

## Turn 1
Prompt: list files in the project
Matched commands: none
Matched tools: LS
Permission denials: 0
stop_reason=completed

## Turn 2
Prompt: list files in the project [turn 2]
Matched commands: none
Matched tools: LS
Permission denials: 0
stop_reason=completed

Use --structured-output when piping turn-loop output into another process that expects JSON, for example a test harness that validates the matched command and tool names.

`remote-mode`

Simulates remote-control runtime branching.

python3 -m src.main remote-mode <target>

target

string

required

Identifier for the remote control target (hostname, URL, or arbitrary label).

Example output

mode=remote
connected=True
detail=Remote control placeholder prepared for prod-machine-01

`ssh-mode`

Simulates SSH runtime branching.

python3 -m src.main ssh-mode <target>

target

string

required

SSH target string (e.g. user@hostname).

Example output

mode=ssh
connected=True
detail=SSH proxy placeholder prepared for user@prod-machine-01

`teleport-mode`

Simulates Teleport runtime branching (resume or create a Teleport session).

python3 -m src.main teleport-mode <target>

target

string

required

Teleport node name or session identifier.

Example output

mode=teleport
connected=True
detail=Teleport resume/create placeholder prepared for teleport-node-42

`direct-connect-mode`

Simulates direct-connect runtime branching.

python3 -m src.main direct-connect-mode <target>

target

string

required

Target label for the direct connection (e.g. a Unix socket path or named pipe).

Example output

mode=direct-connect
target=/tmp/claw.sock
active=True

`deep-link-mode`

Simulates deep-link runtime branching.

python3 -m src.main deep-link-mode <target>

target

string

required

Deep-link URI or target identifier (e.g. claude://open?session=abc123).

Example output

mode=deep-link
target=claude://open?session=abc123
active=True

Remote mode output format

remote-mode, ssh-mode, and teleport-mode return a RuntimeModeReport with three fields:

Field	Type	Description
`mode`	string	One of `remote`, `ssh`, `teleport`
`connected`	boolean	Always `True` in the current placeholder implementation
`detail`	string	Human-readable description of the prepared connection

direct-connect-mode and deep-link-mode return a DirectModeReport:

Field	Type	Description
`mode`	string	One of `direct-connect`, `deep-link`
`target`	string	The target argument as provided
`active`	boolean	Always `True` in the current placeholder implementation

All five remote mode commands are placeholder implementations. They return canned responses and do not establish real connections. They exist to mirror the mode-routing branch structure of the TypeScript runtime.

CLI Commands

Python API

`bootstrap`

Arguments

Output sections

Example

`turn-loop`

Arguments

Per-turn output format

Example

`remote-mode`

`ssh-mode`

`teleport-mode`

`direct-connect-mode`

`deep-link-mode`

Remote mode output format

Build docs developers (and LLMs) love

CLI Commands

Python API

​bootstrap

​Arguments

​Output sections

​Example

​turn-loop

​Arguments

​Per-turn output format

​Example

​remote-mode

​ssh-mode

​teleport-mode

​direct-connect-mode

​deep-link-mode

​Remote mode output format

Build docs developers (and LLMs) love

`bootstrap`

Arguments

Output sections

Example

`turn-loop`

Arguments

Per-turn output format

Example

`remote-mode`

`ssh-mode`

`teleport-mode`

`direct-connect-mode`

`deep-link-mode`

Remote mode output format