Skip to content

chore: release 0.6.1 — scanner-core 0.7.1 - #14

Merged
tiltom merged 2 commits into
mainfrom
chore/0.6.1
Nov 19, 2026
Merged

tiltom merged 2 commits into
mainfrom
chore/0.6.1

Conversation

@tiltom

@tiltom tiltom commented Nov 19, 2026

Copy link
Copy Markdown
Member

Picks up @flagrix/scanner-core 0.7.1. The CLI bundles the engine into its
dist, so its users have had neither the speedup nor the false-positive fixes
until now.

repo before after
facebook/react 140.2s 16.5s 8.5x
expressjs/express 58.9s 5.9s 10.1x

And two false positives on legitimate build code are gone: a bundler's own test
suite read as a build script (which made evanw/esbuild verdict HIGH and
unsafe to clone), and new Function('module', 'exports', 'require', src) — the
CommonJS module wrapper — reported as a critical backdoor.

A patch, not a minor

Both engine fixes only ever remove findings, and the concurrency change is
verdict-identical by construction and by measurement. So no consumer's passing
scan starts failing; some that were failing on a false positive start passing.
That is a bug fix, unlike the exit-code 4 contract change in 0.4.0.

A release bug caught by the smoke test

src/cli.ts carried a hardcoded const VERSION = "0.6.0", which npm version
does not reach — so this bump produced a CLI that reported the previous
release:

built --version: 0.6.0      # after bumping package.json to 0.6.1

It has been kept in step by hand through every release so far (checked v0.4.0
through v0.6.0 — all four matched), so the discipline held; this would have been
the first to break it. release.yml already compares the tag against
package.json and both server.json versions, but never against the literal the
CLI actually prints, and the smoke test only checks that --version runs.

tests/version.test.ts now pins all four copies — package.json, the CLI
literal, server.json .version and server.json packages[0].version — so the
next bump cannot drift.

Verified

  • all three version fields plus the lock at 0.6.1, checked with release.yml's
    own comparison logic run locally
  • node dist/cli.js --version prints 0.6.1
  • typecheck clean, 67 tests pass, build and CLI smoke test ok
  • the engine is the published 0.7.1 from the registry, not a symlinked checkout
  • the full 114-sample benchmark with --enforce-gates run locally — the same
    run release.yml now performs before publishing

This is the first release where a red benchmark blocks the publish, so the gate
and the FLAGRIX_GITHUB_TOKEN secret both get their first real exercise here.
Expect one non-fatal line in that step: 2 completed sample(s) missed their expected verdict, from the two Docusaurus false alarms that belong to
HARDCODED_API_KEY's precision in flagrix-detection-rules.

Picks up @flagrix/scanner-core 0.7.1, which reads a repo's files concurrently
(facebook/react 140.2s -> 16.5s, expressjs/express 58.9s -> 5.9s) and stops
two false positives on legitimate build code: a bundler's own test suite read
as a build script, and `new Function('module', 'exports', 'require', src)`
reported as a critical backdoor. The CLI bundles the engine, so its users had
neither until this release.

A patch, not a minor: both engine fixes only ever remove findings and the
concurrency change is verdict-identical, so no consumer's passing scan starts
failing. Some that were failing on a false positive start passing.

src/cli.ts carried a hardcoded `const VERSION`, which `npm version` does not
reach, so this bump built a CLI that reported 0.6.0. It has been kept in step
by hand through every release so far; tests/version.test.ts now pins it, along
with the two copies in server.json, against package.json. Release CI already
compared the tag with package.json and server.json but never with the literal
the CLI actually prints.
The snapshot being replaced was recorded at engine 0.5.0 on 2026-10-21 — two
engine releases back, not one. That is why 0.7.0's false positives on
evanw/esbuild and reduxjs/redux went unnoticed for a month: there was nothing
current to compare a run against.

This run is the full 114 samples at engine 0.7.1 and CLI 0.6.1, all five gates
green, engineDirty false — the published artifact from the registry rather than
a symlinked checkout, as run.mjs requires of the metadata.

112 of 114 samples pass. The two that do not are reduxjs/redux and
puppeteer/puppeteer, where a Docusaurus config's Algolia DocSearch key reads as
a hardcoded secret. They are deliberately NOT recorded in reference-review.json
as accepted false alarms, because they are not accepted: HARDCODED_API_KEY's
precision is an open question for flagrix-detection-rules. Leaving them as a
standing warning keeps that visible on every run.
@tiltom
tiltom merged commit 562c67b into main Nov 19, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant