hd-bug-hunter-enhanced - 🛡️🔍 Bug Hunter Enhanced | Production-grade vulnera - 剑庭安 - 网站开发定制 SEO 优化蜘蛛池代做在线看手相算命

项目说明

bug-hunter-enhanced

AI-augmented security triage agent for authorized testing only.

bug-hunter-enhanced wraps the upstream codexstar69/bug-hunter reconnaissance/scan toolkit with an autonomous AI reasoning layer that:

Ingests raw scan output (nuclei, ZAP, semgrep, pip-audit, npm-audit, generic formats)

Classifies each candidate (CWE/C mapping)

Filters false positives with CodeT5+ code inspection and BGE retrieval

Computes CVSS v3.1 scores with deduplication and ranking

Generates professional, cited security reports with reproduction + remediation

Learns from ongoing security research (ArXiv cs.CR/cs.SE, NVD/C, Semantic Scholar)

Safety first: Every run requires an explicit authorized scope. Out-of-scope targets are refused before any processing. The agent ne findings; defensi only.

---

Quick Start

CLI

Docker

---

Architecture

Modules

scope_guard.py — Authorization gate (scope parsing, in-scope decision, audit trail)

recon_scanner.py — Normalizer for nuclei/ZAP/semgrep/pip-audit/npm-audit/generic formats

.py — LLM-powered triage with CodeT5+ code inspection and BGE retrieval

se.py — CVSS v3.1 scoring, deduplication, exploitability ranking

report_generator.py — Cited report generation with deterministic Markdown fallback

impro.py — Research agent that crawls cs.CR/NVD/CWE to propose detection rules

benchmark_runner.py — Quality gate (FP-rate, recall, CWE-co baseline)

Tools

llm_client.py — Claude (primary) → OpenAI → Ollama (privacy mode) with graceful fallback

hf_model_manager.py — CodeT5+, BGE-large, BGE-reranker, BART-CNN with heuristic fallbacks

knowledge_updater.py — ArXi Scholar + NVD/C for SECOND-KNOWLEDGE-BRAIN.md

baseline_runner.py — Record upstream baseline on OWASP Juice Shop / DVWA

lab_tuner.py — Tune FP heuristics from authorized lab benchmark data

c.py — Shared C across the agent library

---

API Endpoints (FastAPI)

| Method | Path | Purpose |
|--------|------|---------|
| POST | /api/v1/triage | Run full triage pipeline (primary cross-agent endpoint) |
| GET | /api/v1/findings/{session_id} | Fetch machine-readable findings for a session |
| GET | /api/v1/report/{session_id} | Markdown report for a session |
| GET | /api/v1/sessions | Recent sessions list |
| POST | /api/v1/benchmark | Score triage against labeled ground-truth |
| GET | /api/v1/proposals | Cited impro proposals |
| POST | /api/v1/knowledge/update | Trigger knowledge-brain crawl |
| GET | /api/v1/audit | Authorization/refusal audit trail (safety) |
| GET | /health, /metrics | Li + Prometheus counters |

Example: POST /api/v1/triage

Response (200):

---

Configuration

En

See config/.en for all options.

Runtime Configuration

config/agent_config.yaml — See file for full settings including:

Triage batch size, confidence floor

C, default per CWE

Knowledge updater schedule (weekly Sunday 02:00 UTC)

Quality gates (FP-rate, recall, CWE-co thresholds)

---

Graceful Degradation

The agent is designed to run with zero external dependencies:

No LLM API keys → heuristic CWE mapper + deterministic reports

No GPU/ML stack → hash embeddings + keyword reranking + code signal heuristics

No network → offline-only mode with cached knowledge brain

No Ollama → skip to next provider or fallback

All 58 tests pass in fully offline mode with no API keys.

---

Safety & Authorization

This agent operates only on explicitly authorized targets.

Every analyze() call requires --scope and --auth-id

Out-of-scope targets are refused before any processing

Full audit trail in audit_log table (who, when, what, refused/allowed)

Reports contain reproduction guidance only

No weaponized payloads, no exploit PoCs

Defensive use only. Use against systems you are not authorized to test is prohibited.

---

Testing

Test scenarios documented in tests/test-scenarios.md.

---

De Status

✅ ALL PHASES (0–7) CODE COMPLETE — Production-ready, open-source release.

Phases 0–6: Core modules, orchestrator, quality gates, HuggingFace integration, LLM client, knowledge pipeline, Docker, testing

Phase 7: Cross-agent API endpoints, shared C, lab tuner

External resource notes (code-complete, require li):

Phase 0 baseline: Requires authorized lab (OWASP Juice Shop / DVWA) for li

Phase 3 CodeT5+ benchmark: GPU enables full e; offline-safe with heuristic fallback

Phase 7 lab tuning: Requires authorized lab artifact + ground-truth labels for FP optimization

See PROJECT-DE.md for full details.

---

License

Apache-2.0 — See LICENSE file for details.

---

Upstream Fork

Built on top of codexstar69/bug-hunter (pinned stable tag). The AI layer consumes upstream scan output as a black-box sidecar — no protocol modifications, upgrade-compatible with new upstream releases.

Impro delta: AI triage + C + cited reporting + research agent on top of upstream recon/scan.

---

Citation

If you use this agent in your research, please cite:

---

Contributing

Contributions welcome! Please read CONTRIBUTING.md (if present) and ensure all tests pass before submitting PRs.

---

Support

For questions, issues, or security concerns, please open an issue on GitHub.

For authorized security testing only. Never use against systems you are not authorized to test.

项目信息

开发语言：Python

Star 数：4

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕🛡️🔍 Bug Hunter Enhanced | Production-grade vulnerability triage and security auditing sidecar. Filters raw scanner noise, maps CWE/CVE matrices, and models zero-weaponization defensive reports via a crawl4ai research loop..展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。

登录后才能参与讨论

下载链接

会员才能下载

登录后开通会员

同分类推荐

暂无同类资源

资源标题	hd-bug-hunter-enhanced - 🛡️🔍 Bug Hunter Enhanced \| Production-grade vulnera
所属分类	防护思路
发布作者	admin
发布时间	2026-10-06 00:15:57
下载权限	免费下载
附件数量	2 个下载链接
浏览次数	424
讨论数量	0
主题状态	正常讨论中

hd-bug-hunter-enhanced - 🛡️🔍 Bug Hunter Enhanced | Production-grade vulnera