Make CI Green Skill for Claude

Make CI Green

Re-run failed Jenkins CI jobs for a brave/brave-core PR. Automatically detects

the failure stage and decides whether to use WIPE_WORKSPACE (for build/infra

failures) or a normal re-run (for test/storybook failures).

When to Use

• Flaky test failures on CI that just need a retry
• Build/infra failures that need a workspace wipe to recover
• After pushing new commits to retrigger only the failing checks
• Batch retriggering when multiple platform checks are failing

Environment Requirements

The following environment variables must be set (e.g., in your .envrc):

export JENKINS_BASE_URL=https://$JENKINS_BASE_URL
export JENKINS_USER=<your-jenkins-username>
export JENKINS_TOKEN=<your-api-token>

Get your API token from $JENKINS_BASE_URL/me/configure.

The Job

Step 1: Parse Arguments

Extract the PR number from the user's input. The PR number is required. Check

for --dry-run flag.

| User says | PR number |

| ------------------------------------ | --------------- |

| /make-ci-green 33936 | 33936 |

| "retry ci for 33936" | 33936 |

| "make ci green on PR 33936" | 33936 |

| "re-run failed jobs 33936 --dry-run" | 33936 (dry run) |

Step 2: Analyze Failures

Run the script in dry-run mode to analyze without triggering:

python3 .claude/skills/make-ci-green/retrigger-ci.py <pr-number> --dry-run --format json

Step 3: Present Findings

Show the user what was found:

• Failing Jenkins checks: check name, failed stage, recommended action

(normal vs WIPE_WORKSPACE), reason

• Test failure analysis (for test stage failures): per-test details

including location, PR correlation, upstream flake verdict, existing issues,

and issue filing suggestions

• Non-Jenkins failures (SonarCloud, Socket Security, etc.): listed but not

actionable by this tool

• Pending checks: still running, listed for awareness
• No failures: report that CI is already green

Example output (build/infra failure):

PR 33936: 1 failing Jenkins check(s)

  [DRY-RUN] continuous-integration/linux-x64/pr-head
       Stage: build
       Action: WIPE_WORKSPACE
       Reason: Pre-test stage failure: "build" -> WIPE_WORKSPACE
       URL: https://$JENKINS_BASE_URL/job/brave-core-build-pr-linux-x64/job/PR-33936/2/

Example output (test failure with analysis):

PR 33936: 1 failing Jenkins check(s)

  [DRY-RUN] continuous-integration/linux-x64/pr-head
       Stage: test_brave_unit_tests
       Action: normal
       Reason: Test/post-build stage failure: "test_brave_unit_tests" -> normal re-run
       URL: https://$JENKINS_BASE_URL/job/brave-core-build-pr-linux-x64/job/PR-33936/2/
       Test Failures (2):

         - BraveWalletServiceTest.GetBalance
           Location: brave
           PR correlation: likely_from_pr (PR modifies files in same directory as test: browser/brave_wallet/)

         - WebUIURLLoaderFactoryTest.RangeRequest
           Location: chromium
           Upstream flake: known_upstream_flake
           PR correlation: likely_unrelated (Chromium test with no related chromium_src overrides in PR)
           >> SUGGEST FILING ISSUE: "Test failure: WebUIURLLoaderFactoryTest.RangeRequest"
           Stack trace (last 20 lines):
             ...

Key decisions shown per test failure:

• likely_from_pr: The PR changes overlap with the test's source — developer

should fix their PR, no issue suggested

• likely_unrelated: The failure appears unrelated to the PR — suggests

filing an issue if none exists

• Existing issue found: Shows link to the existing issue instead of

suggesting a new one

• Upstream flake (Chromium tests only): Shows LUCI Analysis verdict for

upstream flakiness

Step 4: Confirm and Trigger

If the user wants to proceed (and not --dry-run), run the script to trigger

rebuilds:

python3 .claude/skills/make-ci-green/retrigger-ci.py <pr-number> --format json

Report the results: which checks were retriggered, the action taken, and any

errors.

WIPE_WORKSPACE Decision Logic

The script examines which pipeline stage failed:

| Failed Stage | Action | Rationale |

| -------------------------------------------------------------------------------------------------------------------------------------- | ------------------ | -------------------------------------------------------- |

| init, checkout, install, config, build, compile, setup, sync, gclient, source, deps, fetch, configure, bootstrap, prepare, environment | WIPE_WORKSPACE | Infrastructure/build failure; workspace may be corrupted |

| storybook, test(s), audit, lint, upload, publish, or anything else | Normal re-run | Test failure likely flaky; no workspace issue |

| Unknown (API error) | Normal re-run | Safe default |

Stage matching is case-insensitive substring matching (e.g., "Build (Debug)"

matches "build").

Test Failure Analysis

When the failed stage is a test stage (not build/infra), the script

automatically:

• Extracts test failures from Jenkins console output (GTest [ FAILED ]

patterns)

• Classifies each test as Brave (src/brave/) or Chromium (src/

excluding brave) using git grep

• Checks upstream flakiness via LUCI Analysis (Chromium tests only —

Brave-specific tests are not in LUCI)

• Correlates with PR changes to determine if the failure is likely caused

by the PR or unrelated

• Searches for existing issues in brave/brave-browser matching the test

name

• Suggests filing an issue only if the failure is likely unrelated to the

PR AND no existing issue exists

PR Correlation Logic

| Scenario | Assessment | Action |

| ----------------------------------------------------- | -------------------- | --------------------------------------------------- |

| PR modifies the test file itself | likely_from_pr | No issue suggestion — developer should fix their PR |

| PR modifies files in the same directory/module | likely_from_pr | No issue suggestion |

| PR has chromium_src/ overrides in related paths | likely_from_pr | No issue suggestion |

| Chromium test with no related chromium_src/ changes | likely_unrelated | Suggest filing issue (if none exists) |

| Brave test with no overlapping PR changes | likely_unrelated | Suggest filing issue (if none exists) |

| Cannot determine (e.g., test source not found) | unknown | No suggestion |

Issue Filing

Issues are never auto-filed. The script only suggests filing and provides:

• Suggested title: Test failure: <TestSuite.TestMethod>
• Body with platform, upstream flake verdict (if applicable), and stack trace
• Suggested label: QA/intermittent
• Target repo: brave/brave-browser

The user must confirm before any issue is created.

Usage Examples

# Dry run: analyze without triggering
python3 .claude/skills/make-ci-green/retrigger-ci.py 33936 --dry-run

# Trigger rebuilds for all failing checks
python3 .claude/skills/make-ci-green/retrigger-ci.py 33936

# JSON output for programmatic use
python3 .claude/skills/make-ci-green/retrigger-ci.py 33936 --format json

# Dry run with JSON output
python3 .claude/skills/make-ci-green/retrigger-ci.py 33936 --dry-run --format json

Exit Codes

• 0: Success (checks found and processed)
• 1: Config error (missing JENKINS_TOKEN)
• 2: API error (GitHub or Jenkins unreachable)
• 3: No failing Jenkins checks found

Limitations

• Only handles Jenkins CI checks (identified by $JENKINS_BASE_URL in the URL)
• Cannot retrigger GitHub Actions or other CI systems (SonarCloud, Socket

Security)

• Requires Jenkins API access with a valid token
• Does not handle PENDING checks (still running)
• WIPE_WORKSPACE detection relies on stage name keyword matching
• Test failure extraction only supports GTest output format ([ FAILED ]

markers)

• Upstream flake check only works for Chromium tests (Brave-specific tests are

not in LUCI)

• PR correlation uses directory-level heuristics; indirect dependencies may not

be detected

• Requires a local chromium src/ checkout for git grep test classification