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
- 脚本__init__.py—
- 参考文档asset.yaml509 B
- 脚本implementation.py564 B
- 参考文档plugin.yaml921 B
03 SKILL.md 正文
Regression Check
Internal ID: S25
Category: verification
Tier: core
Output: RegressionResult
Activate when
- The workflow reaches the
verificationstage and the required inputs are present.
Applicable roles
- Verifier
Required inputs
candidate_patchtest_planbaseline_evidence
Optional inputs
repository_rules
Procedure
- Run the target reproduction and focused tests on the patched state.
- Compare failures with the baseline and separate pre-existing failures.
- Run broader tests only when changed surface or repository rules justify them.
- 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_testshell_execinspect_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
v0001
暂无简介
★ 61
v0001
暂无简介
★ 61
v0001
暂无简介
★ 61
trajectory-to-lesson
Extract a bounded transferable procedure or anti-pattern from a completed trajectory.
★ 61
test-output-interpretation
Turn noisy command output into a localized failure signature without overclaiming.
★ 61
symbol-localization
Rank functions, classes, methods, or configuration blocks inside selected files.
★ 61