Developer Tools

Tells flaky tests from broken ones and quarantines the flaky ones so they stop blocking merges.

Scores failure rate and flip rate; quarantine posts a GitHub commit status from an outbox worker

Confirms a test is flaky, bisects it across recent commits, and names the likely cause.

Each commit runs in a disposable git worktree, so your checkout is never touched

Shows what each container image layer adds, and the bytes later layers hide.

Pulls straight from the registry API, no Docker daemon: 18 MB peak memory on python:3.12-slim

Finds which of your projects ship a package version, and the dependency chain that pulls it in.

Chains matched npm's own tree for all 4,218 packages across five real lockfiles

Catches hardcoded API keys and private keys before they reach git, as a pre-commit hook.

Self-hosted webhook relay: capture, inspect, verify and fan out webhooks with retries.

Verifies GitHub and Stripe signatures; failed deliveries retry from 10s out to 2h

Warns before a site's TLS certificate expires, with Nagios-compatible exit codes.

Verifies chain and hostname, and still reports the date on an already-expired cert

Checks that backup archives aren't silently corrupt, and that backups are still running.

Reads every entry through, checking zip CRC-32s and the gzip trailer the way a restore would

Breaks LLM server latency into queueing, prefill and decode, under realistic load.

Open-loop Poisson arrivals, so a slowing server can't throttle its own load test

Scores how well a RAG answer is grounded, claim by claim, with the evidence cited.

Each claim is judged supported, contradicted or not-enough-info against its top chunks

Routes to an expensive model only when the cheap model's answer looks unsure.

Confidence from token logprobs: mean, entropy or worst token

Reindexes only the RAG chunks that changed, and invalidates only the cached answers they fed.

Cache entries are keyed to the content hashes of the chunks behind each answer

Inspects a prompt, explains what's weak or wasteful, and rewrites it.

Shows what changed between two versions of a terms of service, and which changes matter.

Every explanation must quote the changed text; quotes not found in it are flagged

Debugs a broken API integration with agents for the request, response, auth and schema.

Finds GitHub issues that are probably duplicates of each other.

Replayed 68 real "Duplicate of" closures: the original lands in the top 10 for 36.7%

Drafts release notes from the git history between two refs.

Conventional Commits are sorted without a model call; only the rest go to the LLM

CLI for creating, versioning and composing Claude skills.

Switches SSH keys, git identity, tokens and env vars between developer profiles in one command.

Secrets sealed with AES-GCM under a PBKDF2-derived passphrase key

Diffs two CSV exports by key, even when the files are bigger than RAM.

External merge sort: two 2M-row files diffed in 12 MB of memory

Ask a SQLite database questions in plain English, and see the SQL it ran.

Read-only enforced by SQLite itself: mode=ro plus an authorizer callback

Finds duplicate, look-alike, burst, blurry and screenshot photos, and sets them aside with undo.

Perceptual hashes in a BK-tree, plus a tiled Laplacian blur score

Re-syncs subtitles to the speech in a video's soundtrack, including mid-file drift.

Every line within 50 ms of the truth on its test set, frame-rate drift included

Finds photos that are near copies of each other: resized, recompressed or lightly edited.

A 63-number light-and-dark layout fingerprint, searched by vector similarity

Ranks the senders cluttering an inbox over IMAP, and which unsubscribe links are safe.

Only RFC 8058 one-click unsubscribe is treated as safe to act on

Finds duplicate files and reports the space you'd reclaim before deleting anything.