test-runner

name	test-runner
description	Prove or benchmark the skill-local Agent Runner.
role	quality-gate

Test Runner

Agent Runner for Bun test context.

Status

• Use for Bun pass/fail, coverage, failure repair, triage, benchmark evidence, and output-contract development.
• Use compact mode for routine Bun test gates.
• Use repair mode for hot-context failures in files already being edited.
• Use triage mode for cold-context failures from broader suites or unopened files.
• Use detail lookup when a repair or triage packet lacks enough context.
• Keep lint, format, and type gates on MCP runners.
• Treat Agent Runner as the default Bun test path.

Owner

• Skill prose: skills/test-runner/SKILL.md.
• Command contract and discovery: skills/test-runner/src/command-contract.ts.
• CLI, parser, result model, and runtime behavior: skills/test-runner/src/test-runner.ts.
• Detail artifact read/write behavior: skills/test-runner/src/test-runner.ts.
• Repo-local front doors: skills/test-runner/package.json#scripts.
• Missing-runtime shell preflight: skills/test-runner/src/test-runner.sh.
• Tests: skills/test-runner/src/test-runner.test.ts.
• Runner Benchmark Harness: skills/test-runner/src/test-runner.benchmark.ts.
• Evidence output: skills/test-runner/var/benchmark-output/.

Commands

• Inspect exact runner usage: cd skills/test-runner, then bun run test-runner --help.
• Inspect benchmark usage: cd skills/test-runner, then bun run test-runner-benchmark --help.
• Prove no-MCP Bun adoption: cd skills/test-runner, then bun run test-runner-benchmark --no-mcp-baseline --local-runner ./src/test-runner.sh --mode fixed-gate --gate-preset bun-no-mcp.
• For runner routing, read context/bun-runner.md.

Verification

• Run bun --filter test-runner-scripts test after runner, command-contract, detail-artifact, or benchmark changes.
• Run bun --filter test-runner-scripts typecheck after TypeScript edits.
• Run the fixed-gate benchmark before changing normal runner guidance.

Workflow

• Start with the plan or benchmark question.
• Choose repair mode when the failing file is already in context.
• Choose triage mode when the failure source may be unopened.
• Use detail lookup with a source-run handle when the packet is too terse.
• Run the focused local runner for Bun test, coverage, repair, or triage context.
• Run the Runner Benchmark Harness before any guidance change.
• Require fixed-gate benchmark evidence before changing normal runner guidance.
• Keep lint and typecheck on current guidance.

Safety

• Pass test-target args only after the runner separator.
• Keep generated evidence under the skill-local output path unless deliberately promoted.
• Keep generated detail under skills/test-runner/var/runner-output/ unless deliberately promoted.
• Do not copy flags, output schemas, parser states, or exit tables into this file.
• Use script help and tests for deterministic behavior.

Next Safe Action

• For a routine Bun test gate, run compact mode.
• For coverage, pass Bun coverage args after --.
• For a failing edited test file, run repair mode and use detail lookup only if needed.
• For a broader failing suite, run triage mode and then narrow with compact or repair mode.
• For future guidance changes, run fixed-gate benchmark evidence first.