ci-flake-classifier

name	ci-flake-classifier
description	Classify CI failures on a PR as flaky or genuine using cross-PR recurrence, rerun detection, and symptom pattern matching. Use when a PR has failing CI checks and you need to know which are real regressions vs known flaky tests.
argument-hint	<PR number> [--deep] [--since Nd]

CI Flake Classifier

Classifies CI failures on a PR as flaky or genuine by layering four signals:

1. Cross-PR check recurrence — same CI job failed on other recent PRs (high)
2. Cross-PR test recurrence — same specific test name failed on other PRs (very high, --deep only)
3. Rerun detection — check failed then passed on same commit SHA (high)
4. Symptom pattern matching — error matches known flaky patterns like Cypress timeouts, ECONNRESET (moderate)

Arguments

$ARGUMENTS — one of:

• A PR number (e.g. 1234) — classify that PR's CI failures
• A PR number with --deep (e.g. 1234 --deep) — also fetch logs from other PRs for test-level matching (slower, stronger signal)
• A PR number with --since Nd (e.g. 1234 --since 14d) — widen the scan window (default: 7d)
• Empty — print usage and stop

If no arguments are provided, print:

Usage: /ci-flake-classifier <PR number> [--deep] [--since Nd]

Examples:
  /ci-flake-classifier 1234
  /ci-flake-classifier 1234 --deep
  /ci-flake-classifier 1234 --since 14d

Prerequisites

• gh CLI — authenticated and available
• python3 — available in PATH

Execution

Step 1: Parse arguments and detect repo

owner=$(gh repo view --json owner --jq '.owner.login')
repo=$(gh repo view --json name --jq '.name')

Step 2: Run the classifier

Build the command from $ARGUMENTS:

python3 scripts/classify-ci-failures.py $ARGUMENTS --repo "$owner/$repo"

If --repo was already provided in $ARGUMENTS, don't add it again.

Capture the JSON output from stdout. Progress messages go to stderr.

Step 3: Interpret and present results

Parse the JSON output. The key fields:

{
  "classifications": [
    {
      "check_name": "...",
      "classification": "flaky | suspected_flaky | external_unknown | genuine | deterministic | unknown",
      "confidence": "certain | high | moderate | low",
      "signals": [{"type": "...", "strength": "...", "detail": "..."}],
      "occurrences": 1,
      "failing_tests": [{"name": "...", "file": "...", "error": "..."}]
    }
  
   
     
         
         
  
      

Step 4: Generate the report

## CI Failure Classification — PR #<number>

**<N> failures classified** — <scanned> PRs scanned (<window>d window)

| Check | Classification | Confidence | Signals |
|-------|---------------|------------|---------|
| <check_name> | ⚠️ Flaky | High | <signal details> |
| <check_name> | ❌ Genuine | High | No recurrence or symptom match |
| <check_name> | ⚠️ External | Low | External CI — logs not accessible |
| <check_name> | 🔧 Deterministic | Certain | Build/lint/type-check (never flaky) |

### Flaky Details
<For each flaky/suspected_flaky, list the signals and any failing test names>

### Genuine Failures
<For each genuine, list the failing tests — these need investigation>

### Recommendation
- <N> flaky — safe to rerun or ignore
- <N> genuine — investigate before merging
- <N> external — check Konflux/Prow UI manually

Classification Legend

Classification	Icon	Meaning	Blocks merge?
flaky	⚠️	Strong signal: recurrence or rerun detected	No
suspected_flaky	⚠️	Moderate signal: symptom pattern match only	No
external_unknown	⚠️	External CI (Konflux/Prow) — logs not accessible	No
genuine	❌	No flaky signals — likely a real regression	Yes
deterministic	🔧	Build/lint/type-check — never flaky	Yes
unknown	❓	GHA check but couldn't extract test details	Yes

Signal Strength

Signal	Strength	What it means
test_recurrence	Very high	Same test name failed on 2+ other PRs
check_recurrence	High	Same CI job failed on 2+ other PRs
rerun_detected	High	Failed then passed on same commit SHA
symptom	Moderate	Error matches known flaky pattern
external_ci	None	External CI system — informational only

flaky requires at least one high/very-high signal. suspected_flaky has moderate signals only.