perf(#6059): make LiveGCFClient poll delay injectable for tests

[fullsend-ai-coder]

Summary

• Add an injectable pollDelay field to LiveGCFClient (defaults to time.After) so tests can bypass real sleeps in operation-polling loops
• Update WaitForOperation, waitForIAMOperation, and waitForCloudRunOperation to use the injectable delay
• Inject an immediate-return delay in 4 tests that exercise polling paths (TestLiveGCFClient_WaitForOperation/completes_on_second_poll and 3 UpdateServiceEnvVars polling sub-tests)

Testing

• go test ./internal/dispatch/gcf/... -v — all tests pass (0.164s total, down from ~8s+ with real sleeps)
• go vet ./internal/dispatch/gcf/... — clean
• gofmt — clean
• scan-secrets — clean

Notes

• Follows the same injectable-function pattern already used by iamRetryDelay in this package
• golangci-lint was not available in the sandbox; pre-commit failed due to sandbox network policy blocking git fetch

---

Closes #6059

Post-script verification

• Branch is not main/master (agent/6059-gcf-injectable-poll-delay)
• Secret scan passed (gitleaks — 483ad9bd31407310940d21003a6cfe1b6b52b5c8..HEAD)
• PR body secret scan passed (gitleaks — no-git)

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 7:51 PM UTC · Completed 8:05 PM UTC

Commit: bc07fa8 · View workflow run →

[codecov]

Codecov Report

❌ Patch coverage is 70.00000% with 3 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
internal/dispatch/gcf/gcp.go	70.00%	3 Missing ⚠️

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

Review

Findings

Low

• [comment-style-consistency] internal/dispatch/gcf/gcp.go:182 — The pollDelay field uses a multi-line doc comment, while the adjacent skipUploadURLCheck test-only field uses an inline comment (// testing only: ...). For consistency within the struct, consider matching the inline comment style.

• [pattern-inconsistency] internal/dispatch/gcf/gcp.go — This PR introduces a struct-field injection pattern (pollDelay) for test time control, while the same file uses a package-level var pattern (iamRetryDelay) for a related concern. The patterns have different signatures and semantics (exponential backoff with attempt count vs. simple time.After replacement), so the divergence is partially inherent. Consider filing a follow-up to harmonize toward the struct-field pattern.

---

Labels: PR modifies GCF dispatch polling infrastructure and is primarily a test performance improvement

[ralphbean]

LGTM.

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 5:58 PM UTC · Completed 6:13 PM UTC

Commit: bc07fa8 · View workflow run →

[fullsend-ai-retro]

Retro: PR #6062 — make LiveGCFClient poll delay injectable for tests

Timeline

Time (UTC)	Event
Aug 10, 19:36	Issue #6059 filed
Aug 10, 19:40	Triage agent completed (~4 min)
Aug 10, 19:50	Code agent completed (~10 min), PR created
Aug 10, 20:05	Review agent approved with 2 low-severity findings
Aug 11, 14:36	Human reviewer (@ralphbean) approved — "LGTM"
Aug 13, 17:56	PR merged

What went well

• Single-pass implementation. The code agent produced a clean, correct implementation on the first try — no fix iterations required. The change follows an established injectable-function pattern (iamRetryDelay) already in the codebase.
• Fast pipeline. Issue-to-PR took ~14 minutes of active agent time (triage + code). Total triage-to-merge was ~3 days, dominated by human review wait time.
• Appropriate review findings. The review agent identified two reasonable low-severity issues: comment style inconsistency between test-only fields, and dual injection patterns (struct field vs package variable) that could be harmonized.
• Review agent matched human. The human reviewer approved without additional findings, confirming the review agent's analysis was adequate for this class of change (small, focused Go performance improvement).

Coverage gap — main finding

The code agent introduced a getPollDelay() helper with a nil-guard default branch (if c.pollDelay != nil { return c.pollDelay }; return time.After). Tests only exercise the injected (non-nil) path, leaving the default path uncovered. Result: 70% patch coverage, below the repo's 80% Codecov threshold. The PR was merged with the codecov/patch check failing.

This is already addressed. #2810 ("Code agent should validate patch coverage meets repo threshold before opening PR") was implemented via PR #6138 and closed today (Aug 13). That PR added coverage verification guidance to AGENTS.md, docs/contributing/go-code.md, and a new check-patch-coverage skill. PR #6062's code agent run (Aug 10) predates the fix, so this is an expected gap.

Evidence for existing open issues

• #5148: The review agent's correctness sub-agent found the untested waitForIAMOperation polling loop but classified it as info severity and filtered it below the low threshold. This PR is a concrete case where info-level severity is insufficient for untested conditional branches.
• #4057: The code agent's tests exercise the injected path of getPollDelay() but not the default (nil) path — a concrete example of tests not exercising the specific code path being changed.
• fullsend-ai/agents#397: The review agent could have estimated aggregate patch coverage and flagged the below-threshold result, but this capability is not yet implemented.

No new proposals

All identified improvement opportunities are tracked by existing open issues. The main finding (missing coverage verification) was addressed by #2810 just today. This retro provides supporting evidence for the existing issues listed above rather than novel findings.