[3.0.3] - 2026-02-08

3.0.14

Patch Changes

• #411 d0cc8b6 Thanks @ofri-peretz! - Ship the JavaScript without tsc's layout.

  Every emitted .js is re-written through esbuild's minifyWhitespace, which removes indentation and line breaks. Across the ecosystem that is 3233 kB -> 2023 kB of shipped JavaScript, a 37% cut; on disk a package install drops about 28%. Indentation alone was ~32% of a compiled rule file.

  This is deliberately NOT minification. Identifiers keep their names, string contents are untouched, and the syntax tree is not rewritten — rule meta (messages, schema, docs URLs) stays byte-identical, which is what the docs site and --print-config read, and a stack trace from inside a rule still names the function it came from. Full mangling would have bought another 4 kB gzipped and cost both.

  Verified against the published artifact: identical lint findings including message IDs, identical rule names, and zero differences across every rule's meta, messages, schema and presets.

• Updated dependencies [7663cfd, d0cc8b6]:

  • @interlace/eslint-devkit@1.10.0

3.0.13

Patch Changes

• #381 74bbf60 Thanks @ofri-peretz! - Load rule modules on demand instead of at plugin load.

  Every plugin barrel used to require all of its rules the moment ESLint loaded the plugin, whether or not your config enabled them. plugin.rules[id] is only ever read for rules a config turns on, so the rest was parse-and-compile cost for code that never ran.

  The published entry now exposes each rule behind a getter, so a rule module is read the first time something asks for it. Measured on a 7-plugin config with 34 rules enabled: 163 rule modules loaded and 251 ms of plugin load, against 34 modules and 8.5 ms — total ESLint wall time 251 ms → 109 ms. On a preset that enables most of a plugin (node-security/recommended, 25 of 37) it is a wash, 72 ms → 65 ms. It is never slower; the win scales with how many plugins you stack and how few of their rules you use.

  Nothing about the plugin API changes. Object.keys(plugin.rules) still lists every rule without loading any of them, repeated reads return the same object, and the ./oxlint sub-export is the same plugin object it always was.

  eslint-plugin-jwt and eslint-plugin-vercel-ai-security also re-export their rule objects as named top-level exports, which cannot be deferred — those two keep loading eagerly.

• #381 74bbf60 Thanks @ofri-peretz! - Declare what we support, load only what we use

  tslib is gone from every package. It was a NON-optional peer of @interlace/eslint-devkit, so all 26 plugins declared it as a dependency to satisfy that peer — 124 kB every consumer installed so twelve require("tslib") calls could resolve. The shipped JavaScript now inlines the TypeScript helpers instead (--importHelpers false on the emit pass that already re-writes it), costing ~9.5 kB in devkit. Zero tslib requires remain anywhere; verified by installing every plugin with no tslib in the tree and loading all 26 with every rule intact.

  eslint-plugin-import-next had a phantom dependency. Its rules require("typescript") at module load, but it was declared in neither dependencies nor peerDependencies — it worked only because something else in the tree happened to install it. A clean install crashed the whole plugin, not just the type-aware rules. typescript is now a required peer, which is what the code actually needs.

  23 "technologies we support" declarations did nothing. Seven plugins listed their target libraries in peerDependenciesMeta with no matching peerDependencies entry, and npm ignores meta for a package that is not declared a peer — verified by installing eslint-plugin-express-security and watching nothing install and nothing warn. eslint-plugin-jwt appeared to support six JWT libraries and formally supported none. All 23 are now real optional peers, matching the convention pg, mongodb, prisma and the other nine already followed:

  plugin	technologies now actually declared
  eslint-plugin-jwt	jsonwebtoken, @nestjs/jwt, express-jwt, jose, jwks-rsa, jwt-decode
  eslint-plugin-lambda-security	@aws-sdk/client-lambda, @middy/core, @middy/http-cors, @middy/http-security-headers, @middy/validator
  eslint-plugin-express-security	express, helmet, cors, csurf, express-rate-limit
  eslint-plugin-nestjs-security	@nestjs/common, @nestjs/throttler, class-validator, class-transformer
  eslint-plugin-vercel-ai-security	ai
  eslint-plugin-maintainability, eslint-plugin-react-features	typescript

  All optional, so nothing is installed on the consumer’s behalf — the declaration is the supported-technology signal, which is exactly what it was meant to be.

  A new gate compares declared dependencies against what the emitted JavaScript actually loads, in both directions: a require with no declaration (works until someone installs cleanly) and a declaration nothing requires (weight every consumer pays). It understands that a dependency may exist to satisfy an optional peer of another dependency, which is why eslint-plugin-import-next legitimately declares oxc-resolver that devkit lazily loads.

• #335 47cde07 Thanks @ofri-peretz! - Fix the ./oxlint subpath export, which pointed at src/oxlint.js — a file no build produces. require('<package>/oxlint') threw MODULE_NOT_FOUND on every published package, while every README documented that exact wiring for oxlint's jsPlugins. The export now points at the build output, dist/src/oxlint.js.

  The path was hardcoded in scripts/generate-oxlint-shims.ts, so the generator rewrote any manual correction back to the broken value on the next drift check — fixed there rather than per package.

  This release also carries npm provenance: the affected packages were last published from a workstation, which has no OIDC token to attest with, so the published tarballs had no attestation. Publishing through the release workflow signs them.

• #335 47cde07 Thanks @ofri-peretz! - Fix SDK peer declarations that npm silently ignored

  Twelve plugins listed their target SDKs under peerDependenciesMeta with {"optional": true} but never declared them in peerDependencies. npm drops any peerDependenciesMeta entry that has no matching peerDependencies key, so the metadata was inert — these packages effectively declared no SDK peer at all. Nothing warned: the failure mode of a dependency you never declared is silence.

  Each SDK now appears in both maps, matching the shape eslint-plugin-pg and eslint-plugin-mongodb-security already use — a supported major range in peerDependencies, optional: true in peerDependenciesMeta:

  Plugin	SDK	Range
  express-security	express	^4.0.0 || ^5.0.0
  	helmet	^6.0.0 || ^7.0.0 || ^8.0.0
  	cors	^2.0.0
  	csurf	^1.0.0
  	express-rate-limit	^5.0.0 || ^6.0.0 || ^7.0.0 || ^8.0.0
  jwt	jsonwebtoken	^8.0.0 || ^9.0.0
  	@nestjs/jwt	^9.0.0 || ^10.0.0 || ^11.0.0
  	express-jwt	^7.0.0 || ^8.0.0
  	jose	^4.0.0 || ^5.0.0 || ^6.0.0
  	jwks-rsa	^3.0.0 || ^4.0.0
  	jwt-decode	^3.0.0 || ^4.0.0
  lambda-security	@aws-sdk/client-lambda	^3.0.0
  	@middy/core	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/http-cors	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/http-security-headers	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/validator	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  maintainability	typescript	>=4.8.4
  nestjs-security	@nestjs/common	^9.0.0 || ^10.0.0 || ^11.0.0
  	@nestjs/throttler	^4.0.0 || ^5.0.0 || ^6.0.0
  	class-validator	^0.14.0 || ^0.15.0
  	class-transformer	^0.5.0
  react-features	typescript	>=4.8.4
  vercel-ai-security	ai	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  jwt-security	same six as jwt	(identical ranges)
  openai-security	openai	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@openai/agents	>=0.1.0 <1.0.0
  anthropic-security	@anthropic-ai/sdk	>=0.1.0 <1.0.0
  	@anthropic-ai/claude-agent-sdk	>=0.1.0 <1.0.0
  gemini-security	@google/genai	^1.0.0 || ^2.0.0
  mcp-sdk-security	@modelcontextprotocol/sdk	^1.0.0

  Ranges were taken from each SDK's real release history, bounded below by the oldest major whose call shape the rules still match and above by the current major. cors, csurf, class-transformer and @aws-sdk/client-lambda have only ever shipped one usable major. The ai range spans v4 because require-max-steps deliberately accepts both the v4 maxSteps option and the v5+ stopWhen form. The two typescript entries reuse the >=4.8.4 bound @interlace/eslint-devkit already declares, since these are the same type-aware-graceful rules behind the same optional TS program.

  Every range admits the version this repo's __compatibility__ specs are actually tested against, so the declaration cannot drift from what CI proves.

  The four SDKs still on 0.x (@openai/agents, both Anthropic packages) use an explicit >=0.1.0 <1.0.0 rather than a caret, because ^0.115.0 resolves to >=0.115.0 <0.116.0 — a range narrow enough to warn on almost every real install. These rules match on call shape and never import the SDK, so the honest constraint is the pre-1.0 line, not a single minor.

  peer-declaration-integrity.test.ts now locks the invariant across every workspace package: a peerDependenciesMeta key with no peerDependencies twin fails the suite and is named in the diff. This class had already been fixed once, in a commit that never merged — nothing went red in its absence, so the bug came back on four newly published packages. A silent failure needs a lock, not review attention.

  Nothing to migrate. Every entry stays optional, so no install adds a package or emits a warning when the SDK is absent. What changes is that a consumer on an unsupported major now gets a peer warning instead of nothing — which was the point of the metadata in the first place.

• Updated dependencies [85e57a7, 74bbf60, e5d31ab, 1fb1cad, d1a3d8c]:

  • @interlace/eslint-devkit@1.8.0

3.0.12

Patch Changes

• #364 86baa02 Thanks @ofri-peretz! - Add the ecosystem and oxlint marks to the README logo row. Each plugin now leads with Interlace -> its ecosystem (node, nestjs, express, react, mongodb, postgresql, mysql, sqlite, prisma, drizzle, knex, typeorm, sequelize, lambda, vercel, jwt) -> oxlint -> ESLint; the generic quality plugins carry the row without an ecosystem mark. README-only change - no rule behaviour is affected. The patch bump is what carries the new README onto npm, which only refreshes a package README on publish.

3.0.11

Patch Changes

• #358 1b8c0df Thanks @ofri-peretz! - Fix SDK peer declarations that npm silently ignored

  Seven plugins listed their target SDKs under peerDependenciesMeta with {"optional": true} but never declared them in peerDependencies. npm drops any peerDependenciesMeta entry that has no matching peerDependencies key, so the metadata was inert — these packages effectively declared no SDK peer at all. Nothing warned: the failure mode of a dependency you never declared is silence.

  Each SDK now appears in both maps, matching the shape eslint-plugin-pg and eslint-plugin-mongodb-security already use — a supported major range in peerDependencies, optional: true in peerDependenciesMeta:

  Plugin	SDK	Range
  express-security	express	^4.0.0 || ^5.0.0
  	helmet	^6.0.0 || ^7.0.0 || ^8.0.0
  	cors	^2.0.0
  	csurf	^1.0.0
  	express-rate-limit	^5.0.0 || ^6.0.0 || ^7.0.0 || ^8.0.0
  jwt	jsonwebtoken	^8.0.0 || ^9.0.0
  	@nestjs/jwt	^9.0.0 || ^10.0.0 || ^11.0.0
  	express-jwt	^7.0.0 || ^8.0.0
  	jose	^4.0.0 || ^5.0.0 || ^6.0.0
  	jwks-rsa	^3.0.0 || ^4.0.0
  	jwt-decode	^3.0.0 || ^4.0.0
  lambda-security	@aws-sdk/client-lambda	^3.0.0
  	@middy/core	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/http-cors	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/http-security-headers	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  	@middy/validator	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0
  maintainability	typescript	>=4.8.4
  nestjs-security	@nestjs/common	^9.0.0 || ^10.0.0 || ^11.0.0
  	@nestjs/throttler	^4.0.0 || ^5.0.0 || ^6.0.0
  	class-validator	^0.14.0 || ^0.15.0
  	class-transformer	^0.5.0
  react-features	typescript	>=4.8.4
  vercel-ai-security	ai	^4.0.0 || ^5.0.0 || ^6.0.0 || ^7.0.0

  Ranges were taken from each SDK's real release history, bounded below by the oldest major whose call shape the rules still match and above by the current major. cors, csurf, class-transformer and @aws-sdk/client-lambda have only ever shipped one usable major. The ai range spans v4 because require-max-steps deliberately accepts both the v4 maxSteps option and the v5+ stopWhen form. The two typescript entries reuse the >=4.8.4 bound @interlace/eslint-devkit already declares, since these are the same type-aware-graceful rules behind the same optional TS program.

  Every range admits the version this repo's __compatibility__ specs are actually tested against, so the declaration cannot drift from what CI proves.

  Nothing to migrate. Every entry stays optional, so no install adds a package or emits a warning when the SDK is absent. What changes is that a consumer on an unsupported major now gets a peer warning instead of nothing — which was the point of the metadata in the first place.

• Updated dependencies [e8e9ee6]:

  • @interlace/eslint-devkit@1.7.0

3.0.10

Patch Changes

• #338 dc25c81 Thanks @ofri-peretz! - Re-publish every package so npm carries the optimised artifact

  No source changed. This is a no-op patch whose entire purpose is to ship the artifact the current build already produces.

  Manifests. scripts and devDependencies are now stripped from every published package.json. Neither can do anything in a consumer’s node_modules — npm never runs one and never installs the other — but they shipped in all 27 manifests, cluttered the npm page, and were read by SCA tools scanning installed manifests. No package declares a lifecycle hook, so nothing observable changes. Every published package is bumped so this applies uniformly rather than to a subset.

  Tarballs. 20 packages were last published before the build pipeline changed and still ship AGENTS.md, CHANGELOG.md, JSDoc in the emitted .js, and the full generated .d.ts tree:

  package	published	rebuilt	saving
  eslint-plugin-react-features	547 kB	320 kB	−227 kB
  eslint-plugin-secure-coding	653 kB	477 kB	−176 kB
  eslint-plugin-conventions	241 kB	116 kB	−125 kB
  eslint-plugin-browser-security	380 kB	291 kB	−89 kB
  eslint-plugin-maintainability	178 kB	116 kB	−62 kB
  eslint-plugin-react-a11y	232 kB	173 kB	−59 kB
  eslint-plugin-reliability	148 kB	90 kB	−58 kB
  eslint-plugin-vercel-ai-security	187 kB	130 kB	−57 kB
  eslint-plugin-operability	90 kB	43 kB	−47 kB
  eslint-plugin-jwt	140 kB	95 kB	−45 kB
  eslint-plugin-modularity	98 kB	58 kB	−40 kB
  eslint-plugin-nestjs-security	122 kB	86 kB	−36 kB
  eslint-plugin-sqlite-security	54 kB	20 kB	−34 kB
  eslint-plugin-sequelize-security	54 kB	21 kB	−34 kB
  eslint-plugin-prisma-security	52 kB	19 kB	−33 kB
  eslint-plugin-mysql-security	52 kB	19 kB	−33 kB
  eslint-plugin-typeorm-security	52 kB	19 kB	−33 kB
  eslint-plugin-drizzle-security	52 kB	19 kB	−33 kB
  eslint-plugin-knex-security	51 kB	19 kB	−32 kB
  eslint-plugin-modernization	45 kB	38 kB	−7 kB

  Those 20 go from 3428 kB to 2169 kB — −36.7%. The remaining 7 were released after the pipeline change and only gain the manifest strip.

  A new check in scripts/check-published-artifacts.ts fails the build if scripts or devDependencies ever reappear in a published manifest, so the strip cannot silently regress.

  The dependency ranges did not need updating: every plugin pins @interlace/eslint-devkit with a caret that 1.6.0 satisfies, verified by a clean install of an unchanged plugin resolving devkit 1.6.0 with zero dependencies and no typescript in the tree.

• Updated dependencies [dc25c81]:

  • @interlace/eslint-devkit@1.6.1

3.0.9

Patch Changes

• #294 659f6dc Thanks @ofri-peretz! - Rewrite description and keywords on every published package for npm search discovery. npm ranks on name, description, and keywords, and the registry only picks up these fields at publish — so this is metadata-only and takes effect for each package on its next release.

  Descriptions now lead with the search phrase. Every one starts ESLint plugin for <the thing you'd search> instead of a brand-first or category-first framing, and names the concrete vulnerabilities the plugin actually detects. Three were corrected while doing so:

  • eslint-plugin-import-next claimed "100x faster no-cycle detection". No 100x measurement exists: CLAIMS.md records 3.1x end-to-end (8x in pure rule execution) on a 5,483-file React codebase, and the highest number in any benchmark result is 54.9x on the synthetic corpus. The description now states the real-codebase figure.
  • eslint-plugin-secure-coding claimed SQL injection, XSS and CSRF coverage — none of which are its rules. It now names what it does detect: LDAP, XPath, XXE, GraphQL and template injection, unsafe deserialization, ReDoS, missing authentication, and PII in logs.
  • eslint-plugin-secure-coding ("89 rules") and eslint-plugin-react-a11y ("37 rules") hard-coded rule counts that had drifted from reality. Counts are generated into interlace-numbers.json; hand-typed copies are removed rather than corrected.

  Keywords now match the vocabulary of the plugins that rank. eslint-plugin-security, eslint-plugin-jsx-a11y, eslint-plugin-n and eslint-plugin-import all carry the eslint / eslintplugin / eslint-plugin trio — six of our packages were missing eslintplugin, and every one now carries all three plus static-analysis, linting and code-quality. Security plugins add sast, appsec and vulnerability; node-security and secure-coding also carry nodesecurity, the exact keyword eslint-plugin-security ranks on. Each plugin gained the CWE identifiers and attack names for what it detects (cwe-78 command injection, cwe-22 path traversal, cwe-89 SQL injection, cwe-79 XSS, cwe-347 JWT algorithm confusion, cwe-352 CSRF, cwe-943 NoSQL injection), and node-security gained the crypto vocabulary it had been missing entirely despite absorbing the crypto rule set (crypto, cryptography, weak-hash, md5, sha1, timing-attack).

  No rule behavior, exports, or configuration changes.

• Updated dependencies [e1cdf83, 659f6dc]:

  • @interlace/eslint-devkit@1.4.3

3.0.8

Patch Changes

• #269 7028fe2 Thanks @ofri-peretz! - docs: dual-logo README header (Interlace mark + ESLint mark side by side) and closing Interlace footer — refreshes the README rendered on npmjs.com. No runtime changes.

• Updated dependencies [7028fe2]:

  • @interlace/eslint-devkit@1.4.2

3.0.7

Patch Changes

• #252 d67e395 Thanks @ofri-peretz! - Fix Codecov badge showing "unknown" — switch from flag to component URL format

3.0.6

Patch Changes

• #225 34ff5a8 Thanks @ofri-peretz! - CI-only: pin all coverage thresholds at 100% (integration target, merges last).

3.0.5

Patch Changes

• #200 02e0baf Thanks @ofri-peretz! - fix: republish recommended preset with the correct plugin namespace

  The published builds of eslint-plugin-maintainability and eslint-plugin-operability shipped a recommended config whose plugin KEY (@interlace/maintainability) did not match its rule PREFIX (@interlace/maintainability/maintainability/… — doubled). ESLint cannot resolve that, so spreading ...configs.recommended throws "could not find plugin" the moment a consumer lints a file — under both ESLint 9 and 10.

  The source was corrected in the 2026-05-16 namespace cleanup (alongside react-features, which has since been republished via other changesets), but these two plugins were never bumped — so npm still serves the broken builds and they are the only two doubled-namespace plugins still unfixed downstream. This republishes them from the corrected source.

  Regression lock: packages/eslint-config-interlace/src/ecosystem-integrity.test.ts loads every plugin's every config preset into a real ESLint instance and fails if any rule→plugin reference cannot be resolved. Run it against the built dist/ in the release pipeline (pre-publish) to also catch a stale-artifact publish — the failure mode that let these two ship broken.

3.0.4

Patch Changes

• #197 ecb8491 Thanks @ofri-peretz! - fix: republish recommended with correctly-namespaced, unscoped rule ids

  eslint-plugin-maintainability@3.0.3 and eslint-plugin-operability@3.0.5 shipped a recommended preset whose rule ids carried a doubled, scoped plugin segment (@interlace/maintainability/maintainability/cognitive-complexity) that no registered plugin key matched. Enabling the preset alongside any other config made ESLint throw at load:

  Could not find plugin "@interlace/maintainability/maintainability".

  (and the equivalent @interlace/operability/operability for operability.)

  The source was already corrected to the bare, unscoped form (maintainability/cognitive-complexity under a maintainability plugin key) but was never republished, so npm still served the broken build. This release ships the corrected build. plugin.meta.name is also fixed to the unscoped eslint-plugin-maintainability (was @interlace/eslint-plugin-maintainability, which drifted from the package name and every other plugin).

  Each plugin is configured on its own — there is no unified config. No rule behaviour changes.

  New regression locks in each plugin's index.test.ts reproduce ESLint's rule-id resolution, pin the plugin name and key as unscoped, and load each recommended preset in a real ESLint instance — failing closed if a scoped or doubly namespaced config could ship again.

Bug Fixes

• align codecov component IDs with full package names (2831b968)

Documentation

• fix changelog header format across all packages (c3a15082)

❤️ Thank You

• Ofri Peretz

[3.0.2] - 2026-02-06

Bug Fixes

• align codecov component names and update docs components (0a59a86c)

❤️ Thank You

• Ofri Peretz

[3.0.1] - 2026-02-02

This was a version bump only for eslint-plugin-maintainability to align it with other projects, there were no code changes.

Changelog

All notable changes to @interlace/eslint-plugin-maintainability will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Documentation

• 📘 Launched new documentation site: eslint.interlace.tools
• 📝 Achieved 100% documentation parity (both .md and .mdx files)

[3.0.0] - 2026-02-02

This was a version bump only for eslint-plugin-maintainability to align it with other projects, there were no code changes.

[1.0.0] - 2026-01-26

Added

• Initial stable release with 12 maintainability rules
• LLM-optimized error messages for AI-assisted development
• 100% test coverage across all rules
• ESLint 9 flat config support
• TypeScript type definitions for all rule options

Rules

Complexity Rules

Code Smell Rules

Clean Code Rules

Presets

• recommended - Balanced maintainability thresholds

SOLID Principles Mapping

Rules are annotated with SOLID principle alignment:

• Single Responsibility: max-lines-per-function, max-lines
• Open/Closed: prefer-early-return