ci-failure-classification

Rules for classifying continuous-integration (CI) failures, which are automated checks that build and test code, using logs and control experiments.

In plain words
What is it for?
They guide installation checks, log inspection, identification of the first failure, and comparisons with an equivalent base or control run.
Why use it?
They help distinguish problems caused by a code change from infrastructure failures, flaky tests, or failures that already existed.

Cursor rule for Codex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/prisma/orm/ci-failure-classification
Clone the repo
git clone --depth 1 https://github.com/prisma/orm

Made for: Codex.

Per session 0 Nothing until a file matches its globs; then the whole rule loads.
When invoked 542 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.00542
Opus 5 $0.00000 $0.00271
Sonnet 5 $0.00000 $0.00108
Haiku 4.5 $0.00000 $0.00054

Measured yesterday against content hash 4a31b8afc456, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

ci-failure-classification scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

.agents/rules/ci-failure-classification.mdc · 37 lines

How it starts

The opening of the file, as written. The whole thing — 37 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CI failure classification

Classify every CI failure from evidence before re-running, "fixing", or blaming it: (a) caused by this PR — fix the code; (b) infra/flake — retry and record the fingerprint; (c) already broken on the base — prove it, route to its own fix.

Before classifying anything

  • pnpm install --frozen-lockfile first. Stale install state fakes both reds and greens: stale dist/ (TS2307 on @internal/*), stale deps (a pin moved on the base is invisible until reinstall), stale fixtures. Install after a rebase, not before.
  • Read the actual job log; save it to a file once and grep the file (running-tests.mdc).
  • Find the first failure. One dead shared connection fails dozens of tests at 0ms — count one event, not N failures.
  • Check the diff can reach the failing package at all.

Control experiments

  • A control proves a failure pre-existing only if it ran the same steps. CI skips heavy steps on inert (docs-only) diffs and still reports green — use a non-inert control diff.
  • When the base never runs the failing workflow (pull_request-triggered), a concurrent trivial PR (e.g. dependabot bump) with the identical signature is the equivalent evidence.
  • Reproduce with the real test file at the real commit; an approximate harness produces the right symptom for the wrong cause.

Known flake fingerprints

  • Different assertions across retry attempts of one test — impossible for a deterministic defect; only the raw log shows it.
  • ECONNRESET / Connection terminated unexpectedly at setup, or one connection error then siblings failing at 0ms — the shared dev database died mid-run.
  • All tests green, run exits 1 on an unhandled rejection — teardown race; classify from the rejection's stack.

A fingerprint match is a hypothesis, not a verdict — and a "flake" recurring on consecutive runs of the same tree is a real defect in a timing costume.

When a gate (not a test) fails

Read the gate's own output. If satisfying it would mean describing changes the PR does not contain, or contorting tense to avoid falsehoods, suspect the gate's comparison base — fix the gate, don't contort.

Read the full file on GitHub · 37 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 37 lines · 0 tokens per session scan A 4a31b8afc456

Subscribe to this mod's changes

ci-failure-classification is a cursor rule published in the GitHub repository prisma/orm (47,579 stars, last pushed today), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 542 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.