What problem does it solve? Manually hunting logic bugs across a large multi-language repository is slow and inconsistent, and naive auto-fixing creates false positives and broken code. This Skill runs a disciplined single-pass find-and-fix workflow that scans code with bug-class static analyzers, validates candidates against intentional-design patterns, and only commits fixes that pass a strict safety bar. ## Core Features & Use Cases - Depth-scaled bug hunting: Choose depth 1-10 to control how hard the analysis hunts, from bare analyzer hits to full-subsystem adversarial review. - 17 addressable repo sections: Covers Python, Go, Dockerfile, shell, GitHub Actions, Helm charts, and Vue/TypeScript with per-section toolchains (ruff, go vet, staticcheck, errcheck, hadolint, shellcheck, actionlint, helm lint, vue-tsc). - Fix-bar safety gate: Every fix must be function-body contained, have bounded blast radius, have test coverage (or match the safe-pattern catalogue), and come from a high-signal analyzer rule; everything else is flagged for human review. - Use Case: Run "fix bugs in operator depth 7" to scan the Go operator with go vet, staticcheck, errcheck, and ineffassign, validate each candidate against the 8-concern gauntlet, commit safe fixes individually, and record flagged candidates in a findings memory file. ## Quick Start Ask the agent to work bugs at a chosen depth and scope, for example: run bug-work on the harness and shared sections at depth 5.
