All loops
AgentsEasy 13 min· claude-sonnet-4-5
Refund pre-approver
Reviews refund requests, recommends approve/deny with reasoning.
Read the full Agents build guide →NOT DEPLOYEDNOT DEPLOYED
0140ms
Trigger
cron(0 7 * * *) fired · every day · 07:00
021100ms
Agent
claude-sonnet-4-5 · in 1600 tok · out 717 tok
0390ms
Tools
fetch-mcp/anthropic:messages.create → 200 OK · 360ms
0430ms
Verify
schema check · pydantic v2 passed
05120ms
Output
human review requested · #approvals
0640ms
Notify
audit log written · runbook link attached
SUCCESS
0%
0 runs
P50
0ms
median
P95
0ms
tail
AVG COST
—
per run
LAST OK
never
—
LAST FAIL
never
none
Latency · last 30 runs0 samples
no runs yet
Latest output · what your users see
Agents · Long-running/ Checkpoint · run 148
Subtasks completed
3 / 5 · avg 42s each
Pending human review
1 · refund > $500 threshold
Next tick
resume in 90s from checkpoint c8f21a
// press Test to run once · Watch live to keep streaming · Deploy to make it real
The problem
support,refunds
The outcome
Refund reviews pile up; reps over-refund out of caution.
Ingredients & skills
Secrets
- ANTHROPIC_API_KEY
Providers
- Anthropic
MCP servers
- fetch-mcp
#agent#automation#claude
How it works
Reviews refund requests, recommends approve/deny with reasoning.
Step 1
1 — Define the loop
Single tool-using loop. Claude decides which MCP tool to call until it has enough context.
Step 2
2 — Bind the locker
The locker contains the Anthropic key plus any provider-specific tokens. Nothing is hardcoded.
Step 3
3 — Schedule or trigger
Either cron, webhook, or on-demand from the Locker UI.
One-line deploy
The button above runs the same command with your saved config. This is the raw CLI form.
bash
npx claudeloops deploy refund-pre-approver