From user feedback to a fixwith your approval in between
Paste a feedback link into Claude Code. It reads the report and the screenshots, finds the code, and sends you a plan. It changes nothing until you say yes.
Read and propose only · Never contacts your users
How it works
The slow part of user feedback is not the fix — it is reading the report, finding the screen in the screenshot, and tracking down the code. Hand that to an agent and keep the decision for yourself.
1.Paste a feedback link
Give Claude Code the console link of any feedback thread — a bug report, a screenshot, a one-line complaint.
2.It reads the whole thread
Message, app version, platform, screenshots, replies and your notes. Then it finds the code and investigates — read-only.
3.It sends you a plan
A summary and a plan land on the thread as an internal note and in your Telegram. Nothing has changed yet.
4.You approve, it fixes
Approve or reject on the thread, with an optional note. Only on approval does it write the fix, verify it and report back.
Guardrails, not trust
An agent that can act on a stranger's message needs limits enforced by the server, not by a prompt. These are.
Approval before action
The agent cannot skip the plan. It waits on your decision, and a rejection stops it.
Never talks to your users
No replies, no emails, no status changes. Your users only ever hear from you.
Your apps only, revocable
A personal token, stored hashed and shown once. It reads only your own feedback and stops working the moment you revoke it.
Set up in a minute
- Create a token under Console → Settings → Agent access.
- Install the Claude Code skill and export the token:
mkdir -p ~/.claude/skills/supportdock-feedback && \
curl -fsSL https://supportdock.io/skills/supportdock-feedback/SKILL.md \
-o ~/.claude/skills/supportdock-feedback/SKILL.md
export SUPPORTDOCK_AGENT_TOKEN=sd_agent_your_tokenUsing another agent? The same flow is three plain HTTP endpoints — see the Agent API.