菜单
Skill for the tester agent. Identify which existing tests are relevant to code changes and what new tests are needed to cover gaps. Load WHENEVER code changes are made — during PR review, before committing, or when planning test coverage.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | regression-identification |
| description | Skill for the tester agent. Identify which existing tests are relevant to code changes and what new tests are needed to cover gaps. Load WHENEVER code changes are made — during PR review, before committing, or when planning test coverage. |
Map exactly what changed using git diff. Categorize each change:
[ADD] — New code added[MOD] — Existing code modified[DEL] — Code removed[REF] — Code restructured (behavior unchanged)[CFG] — Configuration/environment changesTrack: which functions/methods were touched, which files, whether interfaces/contracts changed, whether data schemas changed.
For every changed file, trace at minimum:
Dependency mapping techniques:
| Technique | How | Best For |
|---|---|---|
| Static grep/search | Search for imports/references | Quick, small-medium projects |
| Build graph traversal | Use build system deps | Large projects |
| Dynamic call tracing | Run coverage tools on existing tests | Precise method-level mapping |
Flag changes with radius > 3 modules as high-risk. Stop tracing at 5+ modules deep (diminishing returns).
For each changed unit:
| Scenario | Check |
|---|---|
| Changed function | Does a test call this with relevant inputs? |
| New function | Is there a new test? |
| Changed behavior | Do tests assert the OLD behavior that changed? |
| Removed code | Were there tests for the removed functionality? |
| Error paths | Are error handling changes tested? |
Run the relevant existing tests and confirm: all previously passing tests still pass, and any test updated for new behavior is correct.
| Pattern | What's Missing | Risk |
|---|---|---|
| New code, no tests | Untested new functionality | 🔴 High |
| Behavior changed, tests unchanged | Tests pass for wrong reasons | 🔴 High |
| Bug fixed, no regression test | Same bug can reappear | 🟠 Medium |
| Refactored code, coverage dropped | Previously tested paths untested | 🟠 Medium |
| Integration point changed, no integration test | Interface mismatch | 🔴 High |
| Dependency upgraded, no integration test | Breaking change undetected | 🟠 Medium |
Prioritize gaps by: business criticality → change complexity → defect probability → test proximity.
| Gap Type | Best Level | Why |
|---|---|---|
| New logic path | Unit | Fast, precise, isolated |
| Changed behavior | Unit + Integration | Verifies behavior AND integration |
| Bug fix | Unit (reproduces the bug) | Prevents regression |
| API change | Contract | Catches consumer/provider mismatch |
Write the test that would have caught the bug before it was introduced.
| Change Type | Regression Risk | What to Test |
|---|---|---|
| Bug fix | 🟠 Medium | Fix + adjacent behavior + regression test |
| Refactor (no behavior change) | 🟢 Low | Run existing; add characterization tests if missing |
| New feature | 🟢 Low for existing | New paths; verify integration points unchanged |
| Interface/API change | 🔴 High | All callers; contract tests; backward compat |
| Dependency upgrade | 🔴 High | Full suite; check changelog for breaking changes |
| Data schema migration | 🔴 Very High | Migration forward + rollback; data integrity |
| Tier | Trigger | Scope | Budget |
|---|---|---|---|
| Tier 1: Smoke | Every commit | 20-30 critical tests | < 2 min |
| Tier 2: Selective | Every PR/merge | Impact-analyzed tests from Steps 1-3 | < 15 min |
| Tier 3: Complete | Merge to main | Full test suite | Nightly |
| Tier 4: Full E2E | Pre-release | E2E + manual exploration | Before release |
For individual changes, apply Tier 2 — selective, impact-analyzed regression.
Skill for the orchestrator agent. Hierarchical journal system with time-based consolidation. Use at session startup to load context, after completing tasks to record results, and at consolidation intervals. Do NOT manage journals manually.
Generic hierarchical journal system for any agent's operational context with time-based consolidation. Use this skill whenever you need to read, write, or consolidate operational journals — including at session startup, after completing tasks, before ending a session, or when reviewing past work. This skill works for ANY agent type (persona, non-persona, implementer, tester, researcher, etc.). Do NOT attempt to manage journal context manually; always consult this skill for all journal operations.
Skill for the implementer agent. Language-agnostic code implementation workflow for features, bugs, refactors, and tests. Load before writing or modifying any code.
Skill for the orchestrator agent. Complete plan lifecycle management — track, verify, and report on plans from creation through completion. Use whenever creating plans, marking them done, checking active vs completed status, or verifying plan integrity. Do NOT manage plans manually; always use the scripts documented here.
Skill for the orchestrator agent. Decision rules for assigning tasks to the correct specialist agent type. Consult before every subagent dispatch. Load at startup.
Skill for the planner agent. Standard template for producing a final plan ready to hand off to a developer. Provides plan type templates, quality gates, and validation scripts. Do NOT hand off raw, unvalidated plans — always run plan-validate.sh first.