What problem does it solve? Running tests at scale can hang or kill an AI coding session when test discovery fans out across too many files, and truncated output or misclassified failures waste debugging time. This Skill provides safe execution patterns for the test_runner tool and bun shell commands so test runs stay bounded, reproducible, and diagnosable. ## Core Features & Use Cases - Scope Safety Guards: Enforces single-source-file limits and a 50-file fan-out budget for test_runner scopes (convention, graph, impact), returning scope_exceeded instead of blocking the session. - Per-File Isolation Loops: Provides bash and PowerShell loops that run one bun --smol process per test file, matching CI behavior for mock-heavy directories like tests/unit/agents and tests/unit/tools. - Failure Classification & Verification: Distinguishes stale assertions, soft regressions, genuine pre-existing failures, and new regressions, with git worktree commands to verify failures against main. - Use Case: After changing several source files, loop over the affected test directories with per-file bun commands, capture output to a file to avoid truncation, then classify any failures before documenting them in the PR body. ## Quick Start Ask the agent to run the tests for the files you changed using per-file bun commands and classify any failures it finds.
