---
skill_id: gf-ops-ci-flake-classifier
evaluation: free
quality: agent-decision-grade
functional: true
price_usd_full: 11.99
price_label: $11.99 / unlock
kind: service
category: Coding
challenge: https://lvlltd.com/api/pay?skill=gf-ops-ci-flake-classifier
outline: https://lvlltd.com/skills/gf-ops-ci-flake-classifier/outline.json
runtime_after_unlock: src/index.js createSkill
standards: x402, ERC-7857
completion_pass: functional-v1
---

# Free sample — CI Flake Classifier Kit

**Skill ID:** `gf-ops-ci-flake-classifier`  
**Unlock:** $11.99 / unlock USDC on Base via x402  
**What you get after pay:** runnable sealed pack (`createSkill().run`) + workbook + templates  

## Summary

Classify CI flakes vs real regressions with quarantine rules, re-run budgets, and root-cause buckets for agent and app pipelines.

Flaky red builds destroy trust in CI. This kit defines flake signals, quarantine policy, re-run limits, and coding-side fixes so real breaks stay red. Free outline + x402 sealed unlock. Not legal advice; no delivery-speed guarantees.

## When to use

- The same tests fail intermittently and the team re-runs until green instead of fixing root causes.
- Agent eval suites or e2e jobs flake under load and block merges unfairly.
- You need a shared classifier so quarantine is temporary and owned, not a permanent ignore list.

## Decision-grade outline (free)

1. Define flake signals: fails on re-run without code change, order-dependent tests, timing/network sensitivity, shared mutable fixtures.
2. Build a classifier decision tree: real regression vs flake vs infra vs dependency outage with required evidence links.
3. Set re-run budgets: max automatic re-runs per job, then hard fail and open a flake ticket with owner.
4. Apply quarantine rules: @flake tag, expiry date, owner, and tracking issue—no silent skips without expiry.
5. Drive fixes by bucket: isolation, determinism seeds, fake clocks, hermetic network, and agent eval stability patterns.

## Capabilities

- Flake signal taxonomy and evidence-based classifier tree
- Automatic re-run budgets that prevent infinite green-seeking
- Time-boxed quarantine policy with owners and expiry
- Root-cause fix playbooks for timing, order, network, and agent-eval flakes

## Artifacts after unlock

- `ci-flake-classifier.md`
- `quarantine-policy.yaml`
- `flake-fix-playbook.md`

## Detailed description

CI that is red for vibes loses authority; CI that is green only after five re-runs loses honesty. The CI Flake Classifier Kit helps coding and platform teams separate real regressions from noise without normalizing permanent skips. It starts with a flake signal taxonomy: failures that disappear on re-run with no code change, order-dependent suites, timing and network sensitivity, and shared mutable fixtures that leak state across tests—including flaky agent evaluation harnesses. A decision tree forces evidence before labeling: logs, commit range, infrastructure status, and dependency health so “flake” is not a synonym for “I don’t want to look.” Re-run budgets cap automatic retries; after the budget, the job stays red and opens a flake ticket with an owner instead of burning CI minutes forever. Quarantine is allowed but strict: tagged tests, expiry dates, owners, and tracking issues—silent permanent ignores are out of policy. Fix playbooks cover isolation, deterministic seeds, fake clocks, hermetic network layers, and patterns that stabilize agent evals under concurrency. Free outline available; sealed pack unlocks via x402 Base USDC. Engineering process guidance only—not legal advice, not a guarantee of faster merges or higher quality, and no claims about shipping velocity. Use it so red means something again.

## Functional runtime (paid)

After verified payment the sealed pack includes:

| Path | Role |
|------|------|
| `src/index.js` | `createSkill`, `run`, `plan`, `checklist` |
| `src/runbook.js` | Outline-driven steps with done_when |
| `src/tools.js` | Tool schemas for agent registries |
| `WORKBOOK.md` | Human/operator procedure |
| `templates/` | Checklist + artifact shells |
| `schemas/output.schema.json` | Result contract |

### Agent snippet (after unlock)

```js
import { createSkill } from "./src/index.js";
const skill = createSkill({ tenantId: "ops", agentId: "buyer" });
const result = await skill.run({
  goal: "Execute CI Flake Classifier Kit for my catalog",
  context: { skill_id: "gf-ops-ci-flake-classifier" },
  dry_run: false,
});
// result.ok, result.steps, result.artifacts
```

## Pay path

1. Free: this sample + outline.json  
2. `GET https://lvlltd.com/api/pay?skill=gf-ops-ci-flake-classifier` → HTTP 402  
3. Transfer `maxAmountRequired` USDC to `payTo` on Base  
4. `POST /api/pay` with `X-PAYMENT: {"txHash":"0x…","skill":"gf-ops-ci-flake-classifier"}`  
5. Write `sealed_pack.files` to disk or use install-unlocked-pack.mjs  
6. Re-redeem anytime with the same verified txHash  

## Honesty

- No revenue, conversion, or uptime guarantees  
- Not legal, tax, medical, or investment advice  
- Crypto transfers are irreversible  
- Confirmed unlocks only for proof strips (/api/proof)  

*Public `/sealed/gf-ops-ci-flake-classifier/` is teaser-only. Full `src/` is paywalled under locked sealed delivery.*