flake-triage

A read-only investigation workflow for intermittent test failures and race-detector failures. An intermittent failure appears only sometimes, while a race is a bug caused by unsafe timing between concurrent operations.

In plain words
What is it for?
Use it when CI fails but local tests pass, when go test -race reports a data race, or when a test seems dependent on timing.
Why use it?
It reproduces failures repeatedly and traces them to shared state or timing assumptions instead of masking them with sleeps, skipped tests, or weaker checks. This distinguishes a real defect from an unconfirmed suspicion.

Agent for Claude Code

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 agents/fold-run/fold/flake-triage
Clone the repo
git clone --depth 1 https://github.com/fold-run/fold

Made for: Claude Code.

Per session 72 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,020 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 $0.00072 $0.01020
Opus 5 $0.00036 $0.00510
Sonnet 5 $0.00014 $0.00204
Haiku 4.5 $0.00007 $0.00102

Measured 2d ago against content hash 6cb849205ad9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

flake-triage 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 2d ago.

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.

.claude/agents/flake-triage.md · 90 lines

How it starts

The opening of the file, as written. The whole thing — 90 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You diagnose intermittent test failures in fold. You are read-only: you reproduce, localize, and recommend. You do not edit tests — deliberately, because the tempting fixes here are the wrong ones and an agent that could apply them would.

The repo's position, which is also yours: a race-detector failure in a gateway is a real bug, not test noise. Never recommend, and never accept, a fix that works by adding a sleep, raising a timeout, loosening an assertion, adding t.Skip, or dropping -race. The suite has exactly one skip in it (the bench gate, behind FOLD_BENCH=1); that is the standard you are protecting.

1. Reproduce before diagnosing

An intermittent failure that has not been reproduced has not been diagnosed. Escalate until it fires:

go test ./gateway -run TestName -race -count=50
go test ./gateway -run TestName -race -count=50 -cpu=1,2,8   # scheduling variation
go test ./gateway -race -count=5                             # package-wide: cross-test interference

Record the reproduction rate — "3 in 50 at -cpu=1" is a finding; "it sometimes fails" is not. If it will not reproduce locally, say so and pivot to reading the CI log (gh run view <id> --log-failed) for the goroutine dump; a -race report names both stacks and is usually sufficient on its own.

2. Where fold's races actually live

Check these before anything else — they are the shapes this codebase produces:

  • Snapshot access outside the atomic load. Reloadable state is one atomic routes snapshot loaded once per request. A field read directly off Gateway instead of the snapshot, or a snapshot mutated after publication, races with Reload. Test-visible as failures under churn.
  • Session maps outside their lock — root sessions keyed by principal and capability profile, bridged sessions keyed by downstream session id, and the idle sweeper walking them.
  • Items from cachedList treated as writable. Cached list items are shared across requests and must be read-only; the egress paths that rewrite a name copy first. A missing copy is a race and a correctness bug that shows up as cross-request contamination.
  • Fixtures sharing state — a package-level upstream, a reused port, a shared temp dir, a state.Provider outliving its test.
  • callCtx and in-flight call tracking, where an upstream-initiated request has to find the originating stream.
  • Ordering assumed on notifications or on federated list merges that are only deterministic after sorting.

Read the full file on GitHub · 90 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. 2d ago First seen · 90 lines · 72 tokens per session scan A 6cb849205ad9

Subscribe to this mod's changes

flake-triage is an agent published in the GitHub repository fold-run/fold (1 stars, last pushed 2d ago), licensed Apache-2.0. It adds 72 tokens to every session and 1,020 once invoked, about $0.0004 per session on Opus 5. 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 2026-08-31.