AGENT SKILL

regression-check

Check target correctness and nearby behavior without defaulting to an expensive full suite.

License 许可
MIT
所属仓库
syr-cn/PluginRSI
最近更新
2026-10-18
Stars
★ 61
Forks
1

01 前言 Front Matter

name
regression-check
description
Check target correctness and nearby behavior without defaulting to an expensive full suite.

02 资源文件 4

03 SKILL.md 正文

Regression Check

Internal ID: S25
Category: verification
Tier: core
Output: RegressionResult

Activate when

  • The workflow reaches the verification stage and the required inputs are present.

Applicable roles

  • Verifier

Required inputs

  • candidate_patch
  • test_plan
  • baseline_evidence

Optional inputs

  • repository_rules

Procedure

  1. Run the target reproduction and focused tests on the patched state.
  2. Compare failures with the baseline and separate pre-existing failures.
  3. Run broader tests only when changed surface or repository rules justify them.
  4. Record exact commands, exit status, salient output, and elapsed time.

Verification

  • Target evidence is fresh.
  • New failures are classified as patch-caused, pre-existing, or unknown.

Stop conditions

  • All planned tests are complete or a blocker is documented.

Constraints

  • Do not claim broad regression safety from a single narrow test.

Allowed tools

  • run_test
  • shell_exec
  • inspect_failure

Provenance

Synthesized from design patterns in agentless, mini-swe-agent, superpowers. This file is a normalized reusable abstraction, not a verbatim copy of an upstream skill.

04 同仓库更多 Skills