build-sweep

name	build-sweep
model	sonnet
description	Produces a verified build/test status report and auto-fixed modules by sweeping all Go modules in a workspace. Use when: "run a sweep", "check all builds", "nightly sweep", "fix all failing modules", "workspace health check".
category	system-health
triggers	["run a sweep","check all builds","nightly sweep","fix all failing modules","workspace health check"]
tier	1
agents	["primary"]
tool_dependencies	["file_system"]
inputs	[{"name":"workspace_path","type":"string","description":"Root directory containing Go modules to sweep","required":true}]
outputs	[{"name":"sweep_report","type":"text","format":"markdown-table","description":"Verified status table of all modules with build/test results, fixes applied, and commit hashes"}]

Build Sweep Skill

Philosophy

A healthy workspace is one where every module compiles and every test passes. The build sweep is not a debugging session — it is a systematic sweep that treats each module as an independent unit, diagnoses failures mechanically, and applies minimal fixes. The sweep must never trust self-reports: every fix is verified by re-running the exact build and test commands that originally failed.

When to Use

• As a scheduled nightly task to catch drift across a multi-module workspace
• Before a release to verify all modules are green
• After a large refactor or dependency update that may have broken multiple modules
• When onboarding to a workspace to establish a health baseline

Workflow

Phase 1: Detect

1. Find all directories containing go.mod under the workspace path (skip vendor/, .git/, node_modules/, .claude/worktrees/)
2. For each module, run go build ./... and go test ./...
3. Collect results into a table: module | build status | test status | error summary

Phase 2: Diagnose and Fix

For each failing module, dispatch a parallel agent (max 5 concurrent) to:

1. Read the failing build/test output
2. Read the relevant source files to diagnose the root cause
3. Implement the minimal fix — do not refactor, do not add features
4. Re-run that module's build and tests to confirm green

Phase 3: Verify

After ALL agents complete, independently verify each fix:

1. Run the tests yourself — do NOT trust agent self-reports of completion
2. Check actual file diffs with git diff
3. Confirm no duplicate imports, wrong formats, or regressions
4. Run the full suite one final time to catch cross-module integration issues

Phase 4: Commit

If all fixes pass verification:

1. Stage only the fixed files (not unrelated changes)
2. Commit each module separately: fix(sweep): [module] — [one-line description]

Phase 5: Report

Output a verified summary table: module | failure type | fix applied | verification status | commit hash

Output

The skill produces:

1. A markdown table summarizing all module statuses
2. Git commits for each fix applied (one per module)
3. A final pass/fail verdict for the entire workspace