Fuzz rules

SkillDev tools

Lets your agent stress-test React linting rules to find crashes, slowdowns, and false positives.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Fuzz rules skill

About this capability

Fuzz React Doctor rules for crashes, slowness, false positives, and mutation-sensitive diagnostics with @react-doctor/fuzz. Use after rule tests pass, when investigating a fuzz finding, or whenever an eval, review, or user report confirms a new false positive.

What this skill tells your AI

The instructions your AI receives, as published by millionco/react-doctor in .agents/skills/fuzz/SKILL.md and read by ahel’s review.

Use packages/fuzz after focused rule tests pass. Read its README.md for harness architecture and corpus setup.

Run the harness

Run the target rule first:

FUZZ_RULE=<rule-id> FUZZ_STRICT=1 FUZZ_ITERATIONS=500 nr fuzz

Useful variants:

nr fuzz
FUZZ_RULE=<rule-id> FUZZ_SEED=42 nr fuzz
FUZZ_INVARIANTS=1 nr fuzz
FUZZ_CORPUS_DIR=<repository-directory> nr fuzz
FUZZ_RULE=<rule-id> FUZZ_PRINT_SILENT=1 nr fuzz

Confirm that the target rule fires. A silent rule only exercises early exits. If it stays silent, add a triggering shape to packages/fuzz/src/snippet-pools.ts and rerun.

Use FUZZ_TAG=<tag> instead of FUZZ_RULE to exercise an entire registry cohort.

Recommended full sweep (e.g. before a rule-batch release), pointing the corpus at real checkouts. Ready-made corpora work on any machine — see the package README "Canonical corpora": bun scripts/sync-fuzz-corpus.ts materializes a pinned 48-repo sample (symlinking the RDE cache when present, cloning otherwise), and bun scripts/build-bench-corpus.ts extracts react-bench's diagnostic-dense RD-health target files:

cd packages/fuzz
bun scripts/sync-fuzz-corpus.ts
FUZZ_INVARIANTS=1 FUZZ_ITERATIONS=100 FUZZ_CORPUS_DIR=tmp/corpus-repos nr fuzz

Run the direct false-positive check when a rule change affects common syntax:

cd packages/fuzz
bun scripts/hunt-false-positives.ts

Triage findings

Reproducers live in packages/fuzz/tmp/fuzz-findings/.

  • Crash: minimize the program, add a no-throw regression test, fix the rule, and replay the seed
  • Slow case: profile the pathological shape, bound the walk, and keep the existing threshold
  • Verdict drop: fix detection that depends on incidental syntax, then add the rule to the robustness gate when appropriate
  • Invariant violation: decide whether the rule should react to the rewrite; fix and test unexpected changes
  • False positive: add the valid program to the rule tests and fuzz regression corpus

Replay one finding:

FUZZ_RULE=<rule-id> FUZZ_SEED=<seed> FUZZ_ITERATIONS=1 nr fuzz

Preserve false positives

For every confirmed false positive:

  1. Add a minimal fixture to packages/fuzz/corpus/regressions/<rule-id>--<slug>.tsx.
  2. Include the rule ID, weakness class, and source in the fixture header.
  3. Add a focused valid case to the rule test suite.
  4. Add a generator snippet when the existing pools cannot produce the weakness.
  5. Run nr -C packages/fuzz test and replay the target rule.

Use a stable weakness name such as library-idiom, control-flow, wrapper-transparency, name-heuristic, alias-guard, cross-file, framework-gating, paren-shape, default-parameter, dynamic-computed, private-member, or copy-tracking.

Project-level dead-code and dependency findings belong in core tests, not the rule fuzzer.

Report results

Record the command, target-rule fire count, findings, replay seeds, corpus fixtures, and generator changes. Pass confirmed implementation findings to rule-validate.

Signals

GitHub stars
15k
Forks
479
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
fuzz-millionco
Source
github.com/millionco/react-doctor