- Home
- Skills
- Agents & Orchestration
- Cuelight Pro: Claude Code Run Report, Cost Guard, PR Gate
More screenshots
Works with the AI Agents you already use
Cuelight Pro: Claude Code Run Report, Cost Guard, PR Gate
- **Cost Guard** ("keep this run under $5"): before starting more agents, Claude checks spend against your budget, asks you at 80% and stops starting new agents when it is used up.
$9
Cuelight Pro: Claude Code Run Report, Cost Guard, PR Gate
Example session with this skill installed
How did this run go? What should I fix before I merge?
- Read your context and instructions
- Compiled the cuelight pro
Summary
The run scores
44 / 100 (At risk, so far): 1 agent failed, 1 stalled, 1 is looping, and 2 agents edited code without testing it. Spend so far is $6.50. The HTML report includes the agent map with each agent's problems and cost marked.
Files written
cuelight-reports/run-report-20261101-1716-demo-che.md
cuelight-reports/run-report-20261101-1716-demo-che.html
cuelight-reports/run-report-20261101-1716-demo-che.json
The report (Markdown version)
Run report: Checkout rewrite with payments
northwind-shop · 2026-11-01 17:02 · still running
Score: 44 / 100 · At risk (so far)
| Agents | Done | Failed | Stalled | Duration | Spend | Cache hit |
|---|---|---|---|---|---|---|
| 13 | 8 | 1 | 1 | 11m 40s | $6.50 | 96% |
Goal: ship checkout v2 (cart service, payment adapter, webhooks) behind a flag. The build wave is in and PR #42 is open; the verify wave is running, the unit tests failed twice and the docs agent is still running. Next: review the failing payment test.
Fix next
Fix the failing check from “Write the unit tests” (Risk). Its last check (npm test) failed after it edited cart.test.ts.
2.
Test adapter.ts before you merge (Risk). “Build the payment adapter” edited it and ran no test, build or lint afterwards.
3.
Check types.ts by hand (Risk). Written by more than one agent: Build the checkout UI, Implement the cart service.
4. Give “Audit the current checkout flow” your project rules (Risk). It ran without CLAUDE.md, payments.md.
5.
Avoid rebuilding the prompt cache (Saves money, est. ~$1.01). 58,600 tokens were written again. A compaction rewrote the whole context; /compact earlier, at a quiet point, costs less.
3 more in the full report.
What the run produced
2 commits, 1 pull request, 1 push; 6 of 8 test/build/lint runs passed.
- PR northwind/shop#42
aab2295feat: checkout v2 cart service, payment adapter and migrationd5e6d9afeat: verify webhook signatures and dedupe by event id
Score breakdown
- Reliability 9/30: 1 agent failed; 1 agent stalled (went quiet); 1 agent repeating the same call (possible loop)
Verification 7/25: 2 agents edited code and ran no test, build or lint afterwards; 1 agent finished with a failing check
- Efficiency 13.9/20: 52% of cache writes were rebuilds of context already sent; 5 of 77 tool calls failed
- Flow 8.4/15: 140 s lost to API errors; 1 file written by more than one agent
- Instructions 6/10: 1 agent ran without the project's instructions
Agents
| Agent | Status | Checked | Time | Cost |
|---|---|---|---|---|
| Write the unit tests | failed | failing | 1m 20s | $0.11 |
| Review the checkout for security issues | stalled | — | 41s | $0.86 |
| Run the end-to-end suite | running (loop) | — | 2m 20s | $0.17 |
| Build the checkout UI | running | unchecked | 6m 23s | $0.11 |
| Update the developer docs | running | — | 27s | $0.01 |
| Design the checkout v2 architecture | completed | — | 2m 40s | $0.92 |
| Build the payment adapter | completed | unchecked | 4m 30s | $0.19 |
| Implement the cart service | completed | checked | 3m 40s | $0.14 |
| Compare payment provider APIs | completed | — | 2m 44s | $0.09 |
| Audit the current checkout flow | completed | — | 1m 58s | $0.02 |
| Handle payment webhooks | completed | checked | 2m 15s | $0.01 |
| Write the database migration | completed | checked | 2m 05s | $0.01 |
| Inventory analytics events | completed | — | 1m 36s | $0.01 |
Made by Cuelight Run Report 1.0.0 from the transcripts on this machine. Savings are estimates from this run's own numbers.
Next steps
- Fix the failing check in "Write the unit tests" (npm test).
- Run the tests that cover adapter.ts before you merge.
- Check types.ts by hand: two agents wrote it.
Shall I run the missing tests now?
Sample run on a made-up demo session ("Checkout rewrite with payments", 13 subagents).
Connects securely to your tools. The creator never sees your data.
About this skill
What it does
Everything Cuelight Pro offers for Claude Code multi-agent runs, in one skill: a scored run report, an agent map, Cost Guard, PR Gate and a status check. Ask in plain words and Claude runs the right one.
What you get
- Run report ("how did this run go?"): a score from 0 to 100 with a plain label (Excellent, Good, Needs attention, At risk) and the reason for every point lost; fix next, ranked risk first, then money, then time, with estimated savings; what the run produced (commits, PRs, tests); comparison with your earlier runs; an HTML report to share and Markdown to paste.
- Agent map ("show me my agents"): who started whom and whose work fed which agent, with each agent's problems ("Never tested", "Loop", "Shared types.ts") and cost marked, and the slowest chain outlined. Click an agent for its details and fixes.
- Cost Guard ("keep this run under $5"): before starting more agents, Claude checks spend against your budget, asks you at 80% and stops starting new agents when it is used up.
- PR Gate: before opening a pull request, Claude checks for failed agents, untested edits, failing checks, files written by two agents and loops, and says exactly what to fix.
- Status ("what are my agents doing?"): which agents need you right now.
Bought separately: Run Report $14 + Cost Guard $9 + PR Gate $9 = $32.
How it works
Claude Code already writes a record of every session and subagent. Cuelight Pro reads those records on your machine and turns them into answers. Nothing to set up: no account, no server, no API key.
Best for
Developers and teams running several Claude Code subagents at once who want to know what happened, what to fix, and what it cost.
Not for
Live, always-on monitoring (use the free Cuelight dashboard plugin for that), or agents other than Claude Code (not yet).
Privacy
Runs on your machine. Reads only Claude Code's own session records, read-only. Makes no network calls. Credential-shaped text is redacted from everything it writes.
- Works with Claude Code only for now.
- Cost Guard and PR Gate are checks Claude runs and follows, not hard locks.
- Money needs your own price file; otherwise cost shows in tokens. Savings are estimates from your run's own numbers.
- A snapshot when you ask, not live monitoring.
- Cannot tell when an agent is waiting on your permission.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 14 days ago
Requires
Permissions
File Scopes