LoopQuest for Dify

Send your workflow's AI output to LoopQuest for gamified human-in-the-loop review, and get a verdict back. Three tools:

• Create review task — submit content for review and keep going (non-blocking, "monitor"). Returns a task id immediately; the verdict arrives later via the signed webhook to your callback_url, or by calling get_task_status.
• Wait for verdict — submit content and block this node until a human decides (a "gate"). Returns the verdict so you can branch on approve vs flag.
• Get task status — poll a task's status / verdict by id.

Two ways to use it: monitor vs gate

• Monitor (non-blocking): use Create review task. Your workflow proceeds immediately; a human reviews a copy for quality in the background. Great for QA, drift and accuracy metrics.
• Gate (blocking): use Wait for verdict. Nothing downstream happens until a person approves. Use it for decisions that must have a human answerable — refunds, sign-offs, sends, deletions.

Why a dedicated Wait tool? A Dify chatflow can't natively pause a turn and resume on an external callback. Wait for verdict solves this by creating a gate task and polling server-side inside the tool call until the human decides (or the fail-closed timeout is applied), so the node itself blocks — no loop wiring needed. Keep Max wait at or below your Dify request timeout, and set a Fail-closed timeout so a missing decision never hangs the flow (it escalates by default).

Configure

When you add the tool in Dify, set:

• API key — a LoopQuest workspace API key (Workspaces → API keys) or your public ingest secret.
• Base URL — your LoopQuest deployment (defaults to https://loopquest.tomphillips.uk).

Use it in a workflow

Monitor: after your LLM node, add Create review task, map the output into Content, pick a Module, and carry on. Optionally set Title, External id and a Callback URL.

Gate: after your LLM node, add Wait for verdict, map the output into Content (or Claim + Source for Grounding). The node blocks until the verdict, then wire an IF/ELSE on the returned approved — true → send the answer, false → send a fallback or escalate.

Test / record a demo

cd packages/dify-plugin
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

# Local debugging against a Dify instance:
#   set REMOTE_INSTALL_KEY + REMOTE_INSTALL_HOST from Dify → Plugins → "Debug plugin"
python -m main          # connects to Dify for live testing

# Or package and install:
#   dify plugin package ./ -o loopquest.difypkg
#   then upload loopquest.difypkg in Dify → Plugins → Install from local

Then in Dify: add the LoopQuest tool, run a workflow that calls Create review task, open the LoopQuest Review tab, approve/flag, and (if you set a callback_url) watch the signed verdict arrive.

API

Backed by the public LoopQuest API — see https://<your-app>/docs and /openapi.json.