[simonvanlierde]

ClusterFuzzLite ran on every pull request that wasn't markdown-only, about 7.5 minutes each: 40s pulling images, 96s building and 5 minutes fuzzing. Most of those PRs don't change anything the harness can reach.

• The trigger now uses paths instead of paths-ignore: app/**, r9cam_models/**, fuzz/**, .clusterfuzzlite/**, pyproject.toml, uv.lock and the workflow itself. All of app/ stays in scope, because fuzz/relay_frames.py imports app.main.
• PRs that only touch tests, scripts, docs, other workflows or the Dockerfile skip the job. It isn't a required check on main, so a skipped run doesn't block merging.
• fuzz-seconds stays at 300, so a PR that does touch the relay path gets the same fuzzing as before.
• CONTRIBUTING.md and ARCHITECTURE.md no longer say it runs on every pull request.

I left out caching: the build step runs docker build itself with no way to add a layer cache, and storing the corpus between runs needs a separate repo and a token.

Tests: none ran locally, and actionlint isn't installed here. This PR changes the workflow file, so it should trigger the job and show the filter works.

[codspeed]

Merging this PR will not alter performance

1 warning

⚠️ 4 benchmarks are dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

✅ 12 untouched benchmarks

---

_{Comparing ci-fuzz-path-filter (1db78f8) with main (97c4312)}