メニュー
Anti-hallucination rules for code security audit. Prevents false positive vulnerability reports by enforcing file verification, code authenticity, and tech stack matching. Must be loaded by all audit agents.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Feature-centric deep incremental security review harness for Git-backed code changes, with autonomous, agents, and hybrid execution modes.
Optional graph-context enrichment for feature-centric incremental security review using CodeGraph or code-review-graph when available.
Pre-report finding verification contract. Re-checks whether each vulnerability has a real external Source, reachable Sink, bypassable or missing sanitization, and practical exploitability before final reporting.
Sink chain deep tracing methodology with graded code output templates (Critical full-chain / High-Medium key-nodes), reverse tracking rules, and per-hop Read verification.
Agent contract templates for R1 and R2+ rounds, including output format, token budget management, truncation defense, and auto-injection prompt templates.
Jalor internal Spring framework audit extension for endpoint operation authorization and service audit logging coverage.
| name | anti-hallucination |
| description | Anti-hallucination rules for code security audit. Prevents false positive vulnerability reports by enforcing file verification, code authenticity, and tech stack matching. Must be loaded by all audit agents. |
文件验证机制，大幅减少误报。所有审计 Agent 必须遵守。
⚠️ Every finding MUST be based on actual code read via tools
✗ Do NOT guess file paths based on "typical project structure"
✗ Do NOT fabricate code snippets from memory
✗ Do NOT report vulnerabilities in files you haven't read
✓ MUST use Read/Glob to verify file exists before reporting
✓ MUST quote actual code from Read tool output
✓ MUST match project's actual tech stack
Core principle: Better to miss a vulnerability than report a false positive.
⚠️ 严禁幻觉行为 - 违反此规则的发现将被视为无效
1. 先验证文件存在，再报告漏洞
✗ 禁止基于"典型项目结构"猜测文件路径
✗ 禁止假设 config/database.py、app/api.py 等文件存在
✓ 必须使用 Read/Glob 工具确认文件存在后才能报告
2. 引用真实代码
✗ 禁止凭记忆或推测编造代码片段
✗ 禁止编造行号
✓ code_snippet 必须来自 Read 工具的实际输出
✓ 行号必须在文件实际行数范围内
3. 匹配项目技术栈
✗ Rust 项目不会有 .py 文件
✗ 前端项目不会有后端数据库配置
✓ 仔细观察识别到的技术栈信息
4. 知识库 ≠ 项目代码
✗ 知识库中的代码示例是通用示例，不是目标项目的代码
✗ 不要因为知识库说"这种模式常见"就假设项目中存在
✓ 必须在实际代码中验证后才能报告漏洞
错误示例 (幻觉来源):
1. 查询 auth_bypass 知识 → 看到 JWT 示例
2. 没有在项目中找到 JWT 代码
3. 仍然报告 "JWT 认证绕过漏洞" ← 这是幻觉！
正确示例:
1. 查询 auth_bypass 知识 → 了解认证绕过的概念
2. 使用 Read 工具读取项目的认证代码
3. 只有**实际看到**有问题的代码才报告漏洞
4. file_path 必须是你**实际读取过**的文件
核心原则: 宁可漏报，不可误报。质量优于数量。
完整防幻觉方法论见: references/core/anti_hallucination.md (使用 Read 工具加载)