Back to Tidb

TiDB Verification Profiles

.agents/skills/tidb-verify-profile/SKILL.md

26.3.41.2 KB
Original Source

TiDB Verification Profiles

Overview

Use this skill to decide how much local validation to run before and after code changes. Policy requirements still come from AGENTS.md; this skill is the execution guide.

Profiles

WIP (coding loop)

Use while still iterating and not claiming the task is complete.

  • Run only the smallest scoped checks that validate the changed behavior.
  • Prefer targeted unit tests (go test -run <TestName> -tags=intest,deadlock).
  • Avoid slow sweeps by default (make lint, package-wide runs, realtikvtest).

Ready (completion gate)

Use when claiming task completion or PR readiness. Mandatory trigger phrases are defined in AGENTS.md -> Quick Decision Matrix.

  1. Map changed paths to required test surfaces via AGENTS.md -> Task -> Validation Matrix.
  2. Run minimum required targeted tests for those surfaces.
  3. If code changed, run make lint.
  4. Follow AGENTS.md -> Agent Output Contract for final reporting.

Heavy (explicitly required)

Use only when scope or user request requires expensive checks.

  • Examples: CI reproduction, broad refactor confidence, change scope requiring RealTiKV.
  • Never run make bazel_lint_changed unless the user explicitly requests it.