pentesting 0.16.2 β 0.16.4
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +0 -107
- package/dist/main.js +1459 -1479
- package/package.json +4 -7
package/README.md
CHANGED
|
@@ -30,113 +30,6 @@ export PENTEST_MODEL="glm-4.7"
|
|
|
30
30
|
|
|
31
31
|
pentesting
|
|
32
32
|
```
|
|
33
|
-
|
|
34
|
-
---
|
|
35
|
-
|
|
36
|
-
## Features
|
|
37
|
-
|
|
38
|
-
### π€ Multi-Agent Architecture
|
|
39
|
-
|
|
40
|
-
Autonomous penetration testing with specialized agents:
|
|
41
|
-
|
|
42
|
-
```
|
|
43
|
-
Orchestrator β Recon β Vuln β Exploit β Post
|
|
44
|
-
β β β β
|
|
45
|
-
Web Infra Report
|
|
46
|
-
```
|
|
47
|
-
|
|
48
|
-
Each agent is a simple `while(true) { think β tool β observe }` loop with:
|
|
49
|
-
- **Specialized prompts** (not code)
|
|
50
|
-
- **Dedicated tool sets**
|
|
51
|
-
- **State slicing** for token efficiency
|
|
52
|
-
|
|
53
|
-
### π― CTF Expert Knowledge
|
|
54
|
-
|
|
55
|
-
Built-in security expertise including:
|
|
56
|
-
- **Essential options**: `nmap -Pn` (never forget)
|
|
57
|
-
- **CVE detection**: Apache 2.4.49 β CVE-2021-41773
|
|
58
|
-
- **Service-specific exploits**: Samba, VSFTPD, MS17-010
|
|
59
|
-
- **Web attack vectors**: SQLi, XSS, SSRF, XXE
|
|
60
|
-
- **AD infrastructure**: BloodHound, CrackMapExec
|
|
61
|
-
|
|
62
|
-
### π Transparent Execution
|
|
63
|
-
|
|
64
|
-
Full visibility into agent decision-making:
|
|
65
|
-
|
|
66
|
-
```
|
|
67
|
-
Orchestrator agentLoop
|
|
68
|
-
β think: "Start with reconnaissance"
|
|
69
|
-
β tool_call: delegate('recon', 'Subnet scan')
|
|
70
|
-
β
|
|
71
|
-
βββΆ RECON agentLoop
|
|
72
|
-
β β tool_call: nmap -Pn 10.10.10.0/24 [confirm β y]
|
|
73
|
-
β β observe: 3 hosts discovered
|
|
74
|
-
β β escalate β: recon β vuln
|
|
75
|
-
β βββΆ return "Apache 2.4.49 found"
|
|
76
|
-
β
|
|
77
|
-
βββΆ VULN agentLoop
|
|
78
|
-
β tool_call: curl --path-as-is ... [review β yes]
|
|
79
|
-
β observe: /etc/passwd exposure confirmed
|
|
80
|
-
βββΆ return "CVE-2021-41773 Critical confirmed"
|
|
81
|
-
```
|
|
82
|
-
|
|
83
|
-
### π‘οΈ Safety First
|
|
84
|
-
|
|
85
|
-
- **Scope enforcement**: Never attack outside approved targets
|
|
86
|
-
- **Approval gates**: `auto` / `confirm` / `review`
|
|
87
|
-
- **Audit logging**: Every action recorded
|
|
88
|
-
- **Authorized users only**: No unnecessary prompt defenses
|
|
89
|
-
|
|
90
|
-
---
|
|
91
|
-
|
|
92
|
-
## TUI Commands
|
|
93
|
-
|
|
94
|
-
```
|
|
95
|
-
/target <cidr> Set engagement scope
|
|
96
|
-
/start Start autonomous pentest
|
|
97
|
-
/findings Show all findings
|
|
98
|
-
/loot Show credentials & sessions
|
|
99
|
-
/state Show current engagement state
|
|
100
|
-
/yolo Toggle auto-approve mode
|
|
101
|
-
/exit Exit session
|
|
102
|
-
/help Show all commands
|
|
103
|
-
```
|
|
104
|
-
|
|
105
|
-
---
|
|
106
|
-
|
|
107
|
-
## Environment
|
|
108
|
-
|
|
109
|
-
| Variable | Description | Default |
|
|
110
|
-
|----------|-------------|---------|
|
|
111
|
-
| `PENTEST_API_KEY` | API key (required) | - |
|
|
112
|
-
| `PENTEST_BASE_URL` | Custom API endpoint | - |
|
|
113
|
-
| `PENTEST_MODEL` | LLM model | `glm-4.7` |
|
|
114
|
-
|
|
115
|
-
---
|
|
116
|
-
|
|
117
|
-
## Architecture
|
|
118
|
-
|
|
119
|
-
```
|
|
120
|
-
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
|
121
|
-
β Orchestrator β
|
|
122
|
-
β "Delegate, don't execute directly" β
|
|
123
|
-
β β’ delegate_to_agent β’ escalate β’ get_state β’ set_scope β
|
|
124
|
-
ββββββββββββββ¬βββββββββββββββββββββββββββββββββββββββββββββββββ
|
|
125
|
-
β
|
|
126
|
-
ββββββββββΌβββββββββ¬βββββββββ¬βββββββββ¬βββββββββ
|
|
127
|
-
β β β β β β
|
|
128
|
-
βΌ βΌ βΌ βΌ βΌ βΌ
|
|
129
|
-
Recon Vuln Exploit Post Web Infra
|
|
130
|
-
"Info" "Verify" "Approved" "Shell" "Web" "AD"
|
|
131
|
-
```
|
|
132
|
-
|
|
133
|
-
**Key principles:**
|
|
134
|
-
1. Agent = `while(true) { think β tool β observe }`
|
|
135
|
-
2. Agent difference = prompt + tool set (not code)
|
|
136
|
-
3. Communication = dangerβ via Orchestrator, β direct call
|
|
137
|
-
4. All prompts get Scope + State injection
|
|
138
|
-
5. Approval = `auto` | `confirm` | `review`
|
|
139
|
-
|
|
140
33
|
---
|
|
141
34
|
|
|
142
35
|
## Issue Report
|