OpenClaw Patch Inventory (PoC)

Purpose

This file records the exact runtime modifications currently used to make the Telegram + OpenClaw reply-end-controls PoC work inside openclawtest.

It is not the final architecture.

It exists so the current PoC does not depend on vague memory or hand-wavy descriptions.

User: openclawtest
OpenClaw install root:
- /home/openclawtest/.nvm/versions/node/v22.22.2/lib/node_modules/openclaw/dist

Purpose:

inject default Telegram inline buttons into assistant replies for PoC validation

Observed patch point:

line containing:
- buildInlineKeyboard(opts.buttons ?? [[{ text: "A. 繼續", callback_data: "rec:continue" }, { text: "B. 就這樣吧，不需要額外處理", callback_data: "rec:stop" }]])

What it changes:

if no explicit Telegram buttons are supplied, PoC default reply-end buttons are attached

Purpose:

handle reply-end callback data
persist callback choice
visibly update buttons
send callback acknowledgement
read stop state on later inbound messages and suppress proactive continuation behavior

Observed patch points:

raw callback debug file path:
- reply-end-debug.log
pre-answer callback debug write:
- [pre-answer] ... data=...
persisted state file path:
- reply-end-controls.json
callback click handling:
- reply-end-controls: ${choice} clicked ...
resolved button states:
- ✅ A. 繼續
- ⏹ B. 就這樣吧，不需要額外處理
callback acknowledgement replies:
- reply-end-controls: continue received
- reply-end-controls: stop received
inbound stop-state behavior injection:
- if stored state for this chat is stop, append a strong system-style instruction before agent execution to suppress proactive continuation

These are written under the OpenClaw state dir:

This inventory supports the next implementation phase:

The purpose is simply to capture what currently makes the PoC work.