fix(#6818): append [skip ci] to GitLab scaffold MR title

[fullsend-ai-coder]

The scaffoldCommitFn in repos.go appended [skip ci] to meta.CommitMsg for GitLab forges but not to meta.PRTitle. On GitLab projects with merged-results pipelines, only the MR title check (from PR #6723) can suppress dispatch — the synthetic merge commit does not carry commit message trailers. This caused every scaffold install MR to trigger the review agent.

Add the matching PRTitle append so the dispatch job's CI_MERGE_REQUEST_TITLE check prevents the review agent from running on scaffold MRs.

---

Closes #6818

Post-script verification

• Branch is not main/master (agent/6818-skip-ci-mr-title)
• Secret scan passed (gitleaks — 1ac1750a661c6ccc170267c8b3919d2597cf7810..HEAD)
• PR body secret scan: N/A (commit body path)

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 7:24 PM UTC · Completed 7:42 PM UTC

Commit: 6287c2d · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $4.79

[fullsend-ai-review]

Risk Assessment: moderate (2/5)

Details

Minimal 1-line production change with strong test coverage (0.50 ratio) on a high-churn file with frequent recent fixes, yielding moderate risk driven primarily by git history volatility despite very low metadata risk. Re-review confirms no signal changes from prior assessment.

Previous run

Risk Assessment: moderate (2/5)

Details

Minimal 1-line production change with strong test coverage (0.50 ratio) on a high-churn file with frequent recent fixes, yielding moderate risk driven primarily by git history volatility despite very low metadata risk. Score unchanged from prior assessment.

Previous run (2)

Risk Assessment: moderate (2/5)

Details

Minimal 1-line production change with strong test coverage (0.50 ratio) on a high-churn file with frequent recent fixes, yielding moderate risk driven primarily by git history volatility despite very low metadata risk.

[fullsend-ai-review]

Looks good to me

Previous run

Review

Findings

Low

• [error-handling-idiom] internal/cli/repos_test.go:1891 — The test discards the error from runReposInstall via _ = err. The comment explains this is intentional because partial failures are expected (e.g., GitLab bot-token setup requires a real client), but using t.Logf("runReposInstall: %v", err) would make the error visible in verbose test output without breaking the test when partial failures occur.
  Remediation: Replace _ = err with t.Logf("runReposInstall: %v", err) so the error is visible in verbose test output.

---

Next steps:

• /fs-fix — agent addresses review findings automatically
• /fs-fix <your instruction> — agent fixes with your specific guidance
• Push commits directly — review re-runs automatically on push
• /fs-fix-stop — disable automatic fix runs for this PR

Previous run (2)

Review

Findings

Low

• [error-handling-idiom] internal/cli/repos_test.go:1881 — Every other call to runReposInstall in this file captures the return value as err (34 occurrences). This test discards it with _ =, breaking the established convention.
  Remediation: Change _ = runReposInstall(...) to err := runReposInstall(...) to keep the test consistent with the rest of the file.

---

Labels: PR fixes GitLab scaffold install MR title to include [skip ci] for dispatch suppression

---

Next steps:

• /fs-fix — agent addresses review findings automatically
• /fs-fix <your instruction> — agent fixes with your specific guidance
• Push commits directly — review re-runs automatically on push
• /fs-fix-stop — disable automatic fix runs for this PR

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 7:44 PM UTC · Completed 7:52 PM UTC

Commit: 6287c2d · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $1.22

[fullsend-ai-coder]

🔧 Fix agent — iteration 1 (bot-triggered)

Addressed the single low-severity finding: changed blank identifier assignment to named error capture in TestRunReposInstall_GitLabPRTitleIncludesSkipCI to match established file convention. Test passes, secret scan clean.

Fixed (1):

1. error-handling-idiom: _ = runReposInstall breaks established convention (internal/cli/repos_test.go): Changed _ = runReposInstall(...) to err := runReposInstall(...) with _ = err to match the error-handling idiom used by all 34 other call sites in repos_test.go

Tests: passed

Next steps:

• /fs-review — request a re-review of the changes
• /fs-fix <your instruction> — run another fix pass with specific guidance
• Push commits directly — review re-runs automatically on push
  _{Updated by fullsend fix agent}

[fullsend-ai-review]

🤖 Review · ❌ Terminated · Started 7:54 PM UTC · Ended 8:13 PM UTC

Commit: c48514a · View workflow run →

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 7:54 PM UTC · Completed 8:13 PM UTC

Commit: c48514a · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $4.50

[fullsend-ai-coder]

🤖 Finished Fix · ✅ Success · Started 8:14 PM UTC · Completed 8:21 PM UTC

Commit: c48514a · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $0.70

[fullsend-ai-coder]

🔧 Fix agent — iteration 2 (bot-triggered)

Replaced _ = err with t.Logf in the GitLab scaffold MR title test to surface partial-failure details in verbose test output, as requested by the reviewer.

Fixed (1):

1. error-handling-idiom: test discards error via _ = err (internal/cli/repos_test.go): Replaced _ = err with t.Logf("runReposInstall: %v", err) so the partial-failure error is visible in verbose test output without breaking the test.

Tests: passed

Next steps:

• /fs-review — request a re-review of the changes
• /fs-fix <your instruction> — run another fix pass with specific guidance
• Push commits directly — review re-runs automatically on push
  _{Updated by fullsend fix agent}

[fullsend-ai-review]

🤖 Finished Review · ✅ Success · Started 8:23 PM UTC · Completed 8:36 PM UTC

Commit: 0ff7c3b · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $2.52

Superseded by updated review

[fullsend-ai-retro]

🤖 Finished Retro · ✅ Success · Started 8:53 PM UTC · Completed 9:04 PM UTC

Commit: 0ff7c3b · View workflow run →

Runtime: claude · Model: opus → claude-opus-4-6 · Effort: high · Cost: $3.94

[fullsend-ai-retro]

Retro: PR #6819 — fix(#6818): append [skip ci] to GitLab scaffold MR title

Timeline: Issue #6818 opened → triage (6 min) → code agent (19 min) → PR #6819 opened → 3 review cycles + 2 fix cycles → human approval → merge. Total wall-clock: ~2 hours. Total agent cost: $13.73.

The code agent produced a correct 1-line production fix (meta.PRTitle += " [skip ci]" in repos.go) with a new test on the first attempt. The production code required no rework.

Review-fix cycling on low-severity findings (65% of total cost)

The review agent found a [low] error-handling-idiom violation in the test: _ = runReposInstall(...) instead of err := runReposInstall(...) (34 other calls in the file use err :=). However, its suggestion was incomplete — it said to capture the error but didn't specify what to do with the value. The fix agent chose _ = err, which the second review correctly flagged as still discarding the error. The second review suggested t.Logf("...", err), which the fix agent applied. The third review approved.

This two-cycle cascade cost $8.94 (2 extra reviews at $4.50 + $2.52, 2 fix runs at $1.22 + $0.70) — 65% of the total PR cost — for a low-severity test convention issue.

Evidence for existing issues (no new proposals needed):

• Filter low-severity findings from review runs in this repository #6822 — This PR is a textbook case for the severity threshold proposed there. Setting REVIEW_FINDING_SEVERITY_THRESHOLD: "medium" would have eliminated both review-fix iterations entirely, saving $8.94 and ~75 minutes.
• Review agent should catch all findings in the first pass to reduce rework cycles #1582 — The first review's incomplete suggestion (capture the error but don't say what to do with it) directly caused the cascade. A complete suggestion in the first pass would have saved one full cycle.
• Review agent should acknowledge prior-run context when evaluating code added per its own suggestions #3516 — The second review re-evaluated code produced by the fix agent in response to the first review's own suggestion, without acknowledging the prior-run context.
• Code agent should study sibling files for established patterns before generating code #2414 / Code agent: scan existing usages of variables within the same file before introducing new ones #2289 — The code agent could have avoided the finding entirely by studying the 34 existing err := runReposInstall(...) patterns in the same file before writing the test.

E2E behaviour test flake

Run 33432918746 failed on commit c48514a with openshell: error: failed to download ...openshell_0.0.116-1_amd64.deb in the agent-settings.feature:40 scenario. The same test passed on the next commit (run 33435582933) with no relevant code change between them — the only diff was replacing _ = err with t.Logf in an unrelated test. This is a transient GitHub Releases CDN failure. See proposal below.

Related existing issues: #3902 covers retry logic for harness resource fetching (config files from raw.githubusercontent.com) but not the openshell binary download from GitHub Releases — these are different code paths. #4076 addresses yanked OpenShell releases (prevention via minimumReleaseAge) but not transient download failures (resilience via retry).

Proposals filed

• Add retry with backoff for openshell binary download in sandbox setup (in fullsend-ai/fullsend)