findings-contract
Defines the compact JSONL finding schema used by all PR specialist agents and the review auditor.
- License 许可
- MIT
- 所属仓库
- aydabd/cacad
- 最近更新
- 2026-11-08
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- findings-contract
- license
- MIT
- description
- Defines the compact JSONL finding schema used by all PR specialist agents and the review auditor.
03 SKILL.md 正文
Finding Output Contract
All reviewer agents must return compact JSONL findings. One finding per line.
{
"id": "SEC-001",
"agent": "security-reviewer",
"severity": "blocking",
"category": "security",
"file": "src/auth/token.ts",
"line": 42,
"title": "Access token is logged",
"evidence": "The changed code writes accessToken to structured logs.",
"recommendation": "Remove the token from logs or mask it before logging.",
"confidence": "high"
}
Required fields
| Field | Values |
|---|---|
id |
Stable per agent run, prefix by category, e.g. SEC-001 |
agent |
Agent name |
severity |
blocking, non_blocking, note |
category |
Review category |
file |
Relative path, or empty string for repo-level finding |
line |
Integer line number or null |
title |
Short finding title |
evidence |
Concrete evidence from diff/file/tool output |
recommendation |
Minimal actionable recommendation |
confidence |
high, medium, low |
Noise rules
- Do not emit low-confidence style opinions.
- Do not repeat findings already covered by another category if obvious.
- Prefer one precise blocking finding over many weak findings.
- A finding is blocking only when the PR should not merge as-is.
04 同仓库更多 Skills
token-optimization
Use zero-install context minimization first, with optional local optimization tooling when available.
testing-strategy
Testing philosophy and strategy. Use this when deciding what to test, naming tests, or identifying corner cases.
testing-patterns
Test implementation patterns. Use this when writing tests — table-driven tests, mocking, fixtures, assertions, and parallel execution.
testing-integration
Integration and E2E testing design. Use this when writing integration tests, E2E tests, or deciding what belongs in each test layer.
test-review
Test review for missing coverage, weak assertions, flaky tests, isolation, fixtures, and regression risk.
test-impact-selection
Select the minimal useful tests to run based on changed files, impacted components, and repository conventions.