buildgit（buildgit 搜索）

buildgit

A CLI for git operations and monitoring Jenkins CI/CD build pipelines.

What this skill does

• replacement for git push command. Instead of git push use buildgit push
• uses git to push changes then expects a Jenkins build to automatically start, then monitors it
• shows Jenkins build status in many different ways
• analyzes build timing, executor capacity, pipeline structure, and queue contention to enable build optimization
• uses Jenkins REST api to monitor and track a Jenkins pipeline build job
• any command unknown to buildgit is delegated to git, e.g. buildgit log would run git log

All paths in this document are relative to this SKILL.md file's directory, not the project root.

How to use

what is the build status

push the staged changes and monitor the build.  fix any errors you find.

Commands

Default one-line format (status --line, push --line, build --line) is: %s #%n id=%c Tests=%t Took %d on %I (%r)

Default threads format (--threads on TTY monitoring) is: [%-14a] %S %g %p %e / %E

Threads placeholders are separate from --format placeholders: %a=agent %S=stage %g=progress-bar %p=percent %e=elapsed %E=estimate %%=literal% Use width/alignment like %14a or %-14a. BUILDGIT_THREADS_FORMAT sets the default when --threads has no explicit format argument.

Interpreting Output

Exit codes:

Build states: as reported by Jenkins

• SUCCESS — build passed, all tests green
• FAILURE — build failed; output includes failed stage and error details
• BUILDING — build is still running
• ABORTED — build was manually cancelled
• UNSTABLE — build completed with test failures or marked warnings (common for flaky tests)
• NOT_BUILT — build was skipped or not executed (e.g., due to upstream failure or explicit skip)
• UNKNOWN — the state could not be determined (e.g., API error, incomplete data)
• QUEUED — build is waiting in the Jenkins queue and hasn't started running yet
• Tests=!err! in line output means Jenkins test-report data could not be retrieved due to a communication failure (network/sandbox/API connectivity), not a test result.

For failures, summarize the failed stage name, error logs, and test failure details for the user. When a failure needs more detail, prefer scripts/buildgit -v status --json for structured failed-test stdout and scripts/buildgit status --console-text <stage> after scripts/buildgit status --list-stages. --console-text <stage> accepts exact, case-insensitive, and unique partial stage matches. If the requested parent stage has no direct log text, it walks descendant substages and emits their logs in pipeline order. Normal build/status output goes to stdout, including queue updates, stage progress, completion summaries, and verbose diagnostics. stderr is reserved for invalid input, Jenkins communication failures, and transient TTY redraw sequences. Full header output is condensed: Trigger: merges cause plus user (Manual by …, SCM change, Timer, Upstream), Commit: includes the subject line when available, and Agent: is a top-level field instead of a boxed Build Info section. For jobs that trigger downstream builds, full output now shows a hierarchical === Test Results === tree with parent and child rows, while one-line output aggregates the Tests= counts across the whole tree. In --json, the top-level test_results totals match the line output and multi-job builds add a breakdown array with one entry per parent/downstream build; missing reports are represented as null counts.

For snapshot mode defaults, scripts/buildgit status prints one-line output by default on both TTY and non-TTY stdout (TTY adds color).

• Monitoring commands (push, build, status -f) print prior-jobs + estimated build time before monitoring starts
• scripts/buildgit status -f --probe-all only applies to multibranch jobs, waits on the top-level job, then switches to the detected job/branch build once Jenkins starts it
• Parallel branches with local stages {} blocks print as Branch->Substage rows under the branch, reuse the branch ║ marker and agent, expose parallel_path plus parent_branch_stage in --json, and --threads now shows the active Branch->Substage row live while follow-mode monitoring is running.

Build reference rules:

• 0 and -0 mean latest/current build
• Negative values (-1, -2) are relative offsets from latest build
• Build reference and -n are mutually exclusive

Console log (--console): On failed builds, buildgit shows a curated error summary by default.

• --console auto — force the default error log display even when test failures are present
• --console N — show last N raw console lines instead of curated summary (useful for troubleshooting)

For agent-safe follow mode, prefer:

• scripts/buildgit status -f --once to follow exactly one build and exit
• scripts/buildgit status -f --probe-all --once when the next multibranch build may start on a different branch than the current checkout

Build Optimization

Use this workflow when the user wants to reduce build time or understand Jenkins contention:

1. Run scripts/buildgit queue to see whether builds are waiting on capacity or blocked by quiet periods.
2. Run scripts/buildgit agents to identify which executor labels are saturated or offline, then scripts/buildgit agents --nodes to see which physical nodes overlap those labels.
3. Run scripts/buildgit timing --tests on a recent successful build to find slow stages, parallel bottlenecks, and expensive test suites.
4. Run scripts/buildgit timing --tests --by-stage to map those suites back to the stages that published them, or scripts/buildgit timing --compare <a> <b> / scripts/buildgit timing -n <N> to measure whether a rebalance actually helped.
5. Run scripts/buildgit pipeline to understand stage ordering, parallel branches, agent-label placement, and per-stage test suite summaries, then use references/build-optimization.md for the full workflow and constraints.

Dynamic Context

To inject live build state into context before reasoning about build issues:

`scripts/buildgit status --json 2>/dev/null`

References

See references/buildgit-setup.md for setup instructions (required tools, fixing permission errors,project setup, needed network access, docker sandbox instructions) See references/reference.md for real-world output examples (push failures, parallel pipelines, progress bars, live follow mode).