ci(e2e): always report status on PRs and merge queue

[ralphbean]

Summary

• Remove paths filters from the pull_request trigger in e2e.yml so the e2e job always runs and reports a status
• Add merge_group trigger so e2e can be used as a required check in the merge queue
• When no e2e-relevant files changed on a PR, the job short-circuits after checking the file list via the GitHub API — no checkout, no Go setup, no Playwright install

This unblocks making e2e a required status check without blocking docs-only or non-code PRs. Push-to-main keeps its path filters since it doesn't gate the merge queue.

Closes #1987

Test plan

• Verify this PR's own e2e check runs and short-circuits (no e2e-relevant files changed)
• Verify a PR with Go changes still runs the full e2e suite

🤖 Generated with Claude Code

[github-actions]

Site preview

Preview: https://2a77d38b-site.fullsend-ai.workers.dev

Commit: 6aef818d039a88fde76133f2cd1d9d8ca3306c41

[fullsend-ai-review]

Review

Findings

Medium

• [protected-path] .github/workflows/e2e.yml — This file is under .github/, a protected path requiring human approval. The PR links to issue ci: conditional checks block merge queue when they don't run #1987 and explains the rationale for modifying the CI workflow. Human approval is required regardless of review agent assessment.

Low

• [action-version-consistency] .github/workflows/e2e.yml:44 — Pre-existing: actions/checkout@v4 is used here while all other workflows in the repo use @v6 or @v6.0.2. Not introduced by this PR, but worth updating as a follow-up.

• [implementation-detail] .github/workflows/e2e.yml:42 — The grep regex duplicating the path-relevance logic creates a maintenance burden with the push trigger's paths: filter. If e2e-relevant paths change, the grep regex must be updated separately. Consider adding a comment cross-referencing the two locations.

Info

• [edge-case] .github/workflows/e2e.yml:36 — The GitHub REST API returns a maximum of 3000 files per PR (across all pages). For PRs exceeding that limit, the file list would be truncated and the short-circuit check could incorrectly skip tests. This is an extreme edge case unlikely to occur in practice.

[rh-hemartin]

At least filter out docs/ and web/

[ralphbean]

@rh-hemartin what do you mean? Do you see some way that those will make it through the filter on line 43?

The idea here is that we always run the e2e gh workflow, but sometimes that workflow exits early if there's nothing meaningful to test in the change set.

With this, we can make the e2e test a required check because it "always runs" even though many times it won't run anything of substance.

[rh-hemartin]

Sorry, I just commented based on the removal of paths, but didn't go far into it to see that there was a filter. Just recheked and I looked just at the top half of my monitor lol.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 5:44 PM UTC · Completed 5:49 PM UTC
Commit: ba204cb · View workflow run →

[fullsend-ai-retro]

Retro: PR #1988 — ci(e2e): always report status on PRs and merge queue

Workflow quality: Good. This was a clean, human-authored CI change that went through one review cycle and merged without code rework.

Timeline

1. Jun 7 14:46 — PR created by ralphbean. Single file changed: .github/workflows/e2e.yml (+25/-15).
2. Jun 7 14:54 — Review agent posted findings: 1 medium (protected path), 2 low (checkout v4 inconsistency, duplicated path logic), 1 info (API pagination edge case). All findings were accurate and appropriately categorized.
3. Jun 8 08:42 — Human reviewer (rh-hemartin) requested changes, thinking path filtering was removed entirely.
4. Jun 8 18:42 — Author explained the in-workflow grep filter handles the same paths.
5. Jun 9 07:10 — Reviewer approved after re-reading the full diff.
6. Jun 9 17:38 — PR merged.

Assessment

• Review quality: Strong. The review agent's findings were all valid with no false positives. The [implementation-detail] finding about duplicated path logic was particularly useful — it identified a real maintenance burden.
• Rework rate: Zero code rework. The human's change request was resolved by explanation, not code changes.
• Token cost: Efficient — single file change, no unnecessary exploration.
• One proposal: The duplicated path logic between the push trigger's paths: filter and the in-workflow grep regex is a maintenance risk worth tracking.

Proposals filed

• e2e workflow: deduplicate path-relevance logic between push paths filter and PR grep (in fullsend-ai/fullsend)