fathom-mode

name	fathom_mode
description	Activate Fathom Mode — deeply understand the user's intent before generating any response. Use when the user says 'fathom mode'. Fathom Mode asks targeted questions to build a complete Intent Graph, then compiles a high-quality structured prompt for downstream generation.
homepage	https://github.com/ma-ziwei/fathom-mode
metadata	{"openclaw":{"emoji":"🔮","install":[{"id":"uv","kind":"uv","package":"fathom-mode[openclaw]","label":"Install Fathom Mode"}]}}

Fathom Mode

IMPORTANT: You MUST use the shell tool to run python3 -m ftg relay commands. Do NOT simulate, roleplay, or generate responses yourself. All responses come from the python3 -m ftg relay CLI — you are only a relay between the user and the fathom process.

When to activate

Activate when the user says "fathom mode".

Step-by-step protocol

1. User triggers "fathom mode"

Respond: "Fathom Mode activated. What would you like to explore?"

Wait for the user's next message (their actual request).

2. Start a fathom session

Run this shell command with the user's request:

python3 -m ftg relay "USER'S REQUEST HERE"

The command returns JSON. Read the action field and follow the rules below.

3. Handle the JSON response

action value	What you MUST do
ask_user	Show the display field to the user. Wait for their answer.
review	Show the display field. Ask if they want to continue, edit, or execute.
execute	The compiled_prompt field contains the final structured prompt. Use it for downstream generation.
stop	Tell the user the session has ended.

4. Continue the session

When the user replies, run:

python3 -m ftg relay --session SESSION_ID "USER'S REPLY"

Use the session_id from the previous JSON response. Repeat step 3.

5. Special user commands

User says	What to run
"fathom"	python3 -m ftg relay --session SESSION_ID "fathom"
"execute"	python3 -m ftg relay --session SESSION_ID "execute"
"stop"	python3 -m ftg relay --session SESSION_ID "stop"

Rules

1. NEVER generate fathom questions yourself. Always run python3 -m ftg relay and use its output.
2. ALWAYS show the display field from the JSON response to the user as-is.
3. Track the session_id across the conversation. Every continuation must include --session.
4. When action is execute, the compiled_prompt is a high-quality structured prompt — use it as input for your next generation task.
5. The fathom_score field (0-100%) indicates understanding depth. You may mention it to the user.

Environment variables

Fathom Mode auto-detects LLM providers from OpenClaw auth profiles. No extra configuration needed.

If needed, these env vars are also supported: GEMINI_API_KEY, OPENAI_API_KEY, ANTHROPIC_API_KEY, DEEPSEEK_API_KEY.