feat: exclude generated files from the review diff by default

.prettierignore

@@ -0,0 +1,7 @@
+# Written by the release flow's update-changelog.sh, not by hand — its
+# formatting is the generator's business, and checking it turns every PR
+# red when a release commit lands unformatted on main
+CHANGELOG.md
+# npm's output format is the source of truth; reformatting a lockfile
+# invites churn against what the tool regenerates
+package-lock.json

AGENTS.md

@@ -23,8 +23,8 @@ src/
   logger.ts                # structured JSON logger — levels, child contexts, lazy props
   github/                  # GitHub I/O: event payload → PrContext, octokit wrappers (diff fetch, review posting)
   openrouter/              # OpenRouter I/O: @openrouter/sdk wrapper, per-attempt deadline, structured-output retry ladder, cost summary
-  diff/                    # pure transforms over parse-diff output
-  context/                 # workspace I/O: conventions file, changed files, import-trace scan, doc-mention scan, priority docs
+  diff/                    # pure transforms over parse-diff output + diff-level exclusion (patterns, gitattributes linguist rules, wildcard safety cap)
+  context/                 # workspace I/O: conventions file, root .gitattributes, changed files, import-trace scan, doc-mention scan, priority docs
   review/                  # pure review logic: finding schema, phases + stage dispatch, prompt, non-finding filter, unknown-file filter, cross-phase merge, path normalization, selection, comment mapping, title similarity, context notes, summary
   orchestrate.ts           # pipeline + createPromptedGenerateFindings — fully testable with stub clients
 ```

CHANGELOG.md

@@ -1,6 +1,5 @@
 # Changelog
 
-
 ## [0.4.0] — 2026-09-05
 
 ### Features

action.yml

@@ -78,6 +78,36 @@ inputs:
       __snapshots__
     required: false
     default: ""
+  diff_exclude_paths:
+    description: >-
+      Comma-separated folder prefixes or globs removed from the review diff
+      before the token budget check — excluded files are listed by name in
+      the review output but their content is not reviewed (excluded is not
+      vetted). Supplied patterns EXTEND a built-in default list of
+      generated artifacts (ecosystem lockfiles, *.min.js, *.min.css,
+      *.map — the classes GitHub's linguist auto-collapses); a leading
+      "none" drops the defaults, so "none" alone disables the built-in
+      list (linguist-generated exclusions are governed separately by
+      respect_linguist_generated) and "none, evals/**" replaces the list
+      outright. Empty = the default
+      list (unlike exclude_paths, where empty means no exclusions), so
+      workflows can wire an unset repo variable directly. Patterns with
+      more than 2 "*" in one path segment are rejected ("**" segments are
+      exempt) — glob matching backtracks exponentially on such shapes.
+      Example: none, **/__snapshots__/**
+    required: false
+    default: ""
+  respect_linguist_generated:
+    description: >-
+      Also exclude changed files the repo's root .gitattributes marks
+      linguist-generated=true (nested .gitattributes files are not read).
+      Negated entries (-linguist-generated) keep a file reviewable even
+      when the default diff_exclude_paths list matches it; explicitly
+      supplied diff_exclude_paths patterns always win. Rules are read from
+      the PR head, so a PR changing .gitattributes reviews under its own
+      rules — exclusions are always named in the review output
+    required: false
+    default: "true"
   cost_summary:
     description: Write a per-run cost report (model, prompt/completion tokens, USD) to the workflow step summary
     required: false

package.json

@@ -34,6 +34,7 @@
     "@actions/github": "9.1.1",
     "@openrouter/sdk": "1.2.85",
     "env-var": "7.5.0",
+    "ignore": "5.3.2",
     "luxon": "3.7.2",
     "parse-diff": "0.12.0",
     "zod": "4.5.4"

src/__tests__/config.test.ts

@@ -1,5 +1,9 @@
 import { describe, expect, it } from "vitest"
-import { parseConfig, type RawInputs } from "../config.js"
+import {
+  DEFAULT_DIFF_EXCLUDE_PATTERNS,
+  parseConfig,
+  type RawInputs,
+} from "../config.js"
 
 const makeRawInputs = (overrides: Partial<RawInputs> = {}): RawInputs => ({
   githubToken: "ghs_testtoken",
@@ -19,6 +23,8 @@ const makeRawInputs = (overrides: Partial<RawInputs> = {}): RawInputs => ({
   maxRelatedDocs: "4",
   priorityDocs: "README.md",
   excludePaths: "",
+  diffExcludePaths: "",
+  respectLinguistGenerated: true,
   costSummary: true,
   prNumberOverride: "",
   ...overrides,
@@ -46,6 +52,28 @@ describe("parseConfig", () => {
       maxRelatedDocs: 4,
       priorityDocs: ["README.md"],
       excludePaths: [],
+      diffExcludePaths: {
+        defaultPatterns: [
+          "**/package-lock.json",
+          "**/npm-shrinkwrap.json",
+          "**/yarn.lock",
+          "**/pnpm-lock.yaml",
+          "**/bun.lock",
+          "**/bun.lockb",
+          "**/deno.lock",
+          "**/composer.lock",
+          "**/Cargo.lock",
+          "**/Gemfile.lock",
+          "**/poetry.lock",
+          "**/uv.lock",
+          "**/go.sum",
+          "**/*.min.js",
+          "**/*.min.css",
+          "**/*.map",
+        ],
+        operatorPatterns: [],
+      },
+      respectLinguistGenerated: true,
       costSummary: true,
       prNumberOverride: undefined,
     })
@@ -230,6 +258,78 @@ describe("parseConfig", () => {
     expect(config.excludePaths).toEqual(["evals", "fixtures", "nested/deep"])
   })
 
+  it("extends the default diff_exclude_paths list with supplied patterns", () => {
+    const config = parseConfig(
+      makeRawInputs({ diffExcludePaths: "evals/**, **/*.snap" }),
+    )
+
+    expect(config.diffExcludePaths.operatorPatterns).toEqual([
+      "evals/**",
+      "**/*.snap",
+    ])
+    // The behavioral claim is tier preservation — supplied patterns must not
+    // replace the built-in list, so identity with the constant is the spec
+    expect(config.diffExcludePaths.defaultPatterns).toEqual(
+      DEFAULT_DIFF_EXCLUDE_PATTERNS,
+    )
+  })
+
+  it("disables the default list with a leading none", () => {
+    const config = parseConfig(makeRawInputs({ diffExcludePaths: "none" }))
+
+    expect(config.diffExcludePaths).toEqual({
+      defaultPatterns: [],
+      operatorPatterns: [],
+    })
+  })
+
+  it("replaces the default list via none followed by patterns", () => {
+    const config = parseConfig(
+      makeRawInputs({ diffExcludePaths: "none, evals/**" }),
+    )
+
+    expect(config.diffExcludePaths).toEqual({
+      defaultPatterns: [],
+      operatorPatterns: ["evals/**"],
+    })
+  })
+
+  it("rejects a non-leading none in diff_exclude_paths", () => {
+    expect(() =>
+      parseConfig(makeRawInputs({ diffExcludePaths: "evals/**, none" })),
+    ).toThrow(
+      'diffExcludePaths: "none" disables the default list only in leading position — move it first or remove it',
+    )
+  })
+
+  it("normalizes diff_exclude_paths pattern spellings", () => {
+    const config = parseConfig(
+      makeRawInputs({ diffExcludePaths: "/evals, ./fixtures/, generated//" }),
+    )
+
+    expect(config.diffExcludePaths.operatorPatterns).toEqual([
+      "evals",
+      "fixtures",
+      "generated",
+    ])
+  })
+
+  it("rejects a diff_exclude_paths pattern over the wildcard cap", () => {
+    expect(() =>
+      parseConfig(makeRawInputs({ diffExcludePaths: "*a*a*a*b" })),
+    ).toThrow(
+      'diffExcludePaths: pattern(s) exceed the wildcard cap (at most 2 "*" per path segment; "**" segments exempt): *a*a*a*b',
+    )
+  })
+
+  it("passes a false respect_linguist_generated through unchanged", () => {
+    const config = parseConfig(
+      makeRawInputs({ respectLinguistGenerated: false }),
+    )
+
+    expect(config.respectLinguistGenerated).toBe(false)
+  })
+
   it("rejects a zero max_related_files", () => {
     expect(() => parseConfig(makeRawInputs({ maxRelatedFiles: "0" }))).toThrow(
       'maxRelatedFiles: "0" is not a positive integer',