clawarmor 3.0.0 → 3.0.1
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 +59 -22
- package/package.json +1 -1
package/README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# ClawArmor
|
|
2
2
|
|
|
3
|
-
|
|
3
|
+
The security control plane for OpenClaw agents — audit, harden, and orchestrate your full protection stack.
|
|
4
4
|
|
|
5
5
|
[](https://www.npmjs.com/package/clawarmor)
|
|
6
6
|
[](LICENSE)
|
|
@@ -8,43 +8,77 @@ Security armor for OpenClaw agents — audit, scan, monitor.
|
|
|
8
8
|
|
|
9
9
|
## What it does
|
|
10
10
|
|
|
11
|
-
|
|
12
|
-
|
|
13
|
-
|
|
11
|
+
AI agent security isn't one tool — it's a stack. ClawArmor is the foundation and control plane:
|
|
12
|
+
|
|
13
|
+
1. **Audits** your OpenClaw config and live gateway — 30+ checks, scored 0–100
|
|
14
|
+
2. **Hardens** your setup — auto-applies safe fixes, snapshots before every change
|
|
15
|
+
3. **Orchestrates** the full security stack — deploys and configures [Invariant Guardrails](https://github.com/invariantlabs-ai/invariant) and [IronCurtain](https://github.com/provos/ironcurtain) based on your audit results
|
|
16
|
+
|
|
17
|
+
```
|
|
18
|
+
clawarmor audit → understand your risk (0–100 score)
|
|
19
|
+
clawarmor stack plan → see what protection stack your risk profile needs
|
|
20
|
+
clawarmor stack deploy → deploy it in one command
|
|
21
|
+
clawarmor stack sync → keep everything aligned after changes
|
|
22
|
+
```
|
|
14
23
|
|
|
15
24
|
## Quick start
|
|
16
25
|
|
|
17
26
|
```bash
|
|
18
27
|
npm install -g clawarmor
|
|
19
|
-
clawarmor protect --install
|
|
20
|
-
clawarmor audit
|
|
28
|
+
clawarmor protect --install # install guard hooks
|
|
29
|
+
clawarmor audit # score your setup
|
|
30
|
+
clawarmor stack deploy --all # deploy full protection stack
|
|
21
31
|
```
|
|
22
32
|
|
|
33
|
+
## The Stack
|
|
34
|
+
|
|
35
|
+
ClawArmor sits at the foundation and orchestrates the layers above it:
|
|
36
|
+
|
|
37
|
+
| Layer | Tool | What it does | ClawArmor role |
|
|
38
|
+
|---|---|---|---|
|
|
39
|
+
| **Foundation** | ClawArmor | Config hygiene, credential checks, skill supply chain | Audits + hardens |
|
|
40
|
+
| **Flow guardrails** | [Invariant](https://github.com/invariantlabs-ai/invariant) | Detects multi-step attack chains at runtime | Generates rules from audit findings |
|
|
41
|
+
| **Runtime sandbox** | [IronCurtain](https://github.com/provos/ironcurtain) | Policy-enforced tool call interception, V8 isolate | Generates constitution from audit findings |
|
|
42
|
+
| **Action gating** | [Latch](https://github.com/latchagent/latch) | Human approval for risky actions via Telegram | Coming in v3.2 |
|
|
43
|
+
|
|
44
|
+
`clawarmor stack deploy` reads your audit score, generates the right config for each tool, and deploys them. `clawarmor stack sync` keeps everything updated as your setup changes.
|
|
45
|
+
|
|
23
46
|
## Commands
|
|
24
47
|
|
|
48
|
+
### Core
|
|
49
|
+
|
|
25
50
|
| Command | Description |
|
|
26
51
|
|---|---|
|
|
27
52
|
| `audit` | Score your OpenClaw config (0–100), live gateway probes, plain-English verdict |
|
|
28
53
|
| `scan` | Scan all installed skill files for malicious code and SKILL.md instructions |
|
|
29
54
|
| `prescan <skill>` | Pre-scan a skill before installing — blocks on CRITICAL findings |
|
|
30
|
-
| `
|
|
31
|
-
| `
|
|
32
|
-
| `protect --status` | Show current protection state |
|
|
33
|
-
| `watch` | Monitor config and skill changes in real time |
|
|
34
|
-
| `watch --daemon` | Start the watcher as a background daemon |
|
|
35
|
-
| `harden` | Interactive hardening wizard (--dry-run, --auto) |
|
|
55
|
+
| `fix` | Auto-apply safe fixes (--dry-run to preview, --apply to run) |
|
|
56
|
+
| `harden` | Interactive hardening wizard (--dry-run, --auto, --monitor) |
|
|
36
57
|
| `status` | One-screen security posture dashboard |
|
|
37
|
-
| `log` | View the audit event log |
|
|
38
|
-
| `digest` | Show weekly security digest |
|
|
39
58
|
| `verify` | Re-run only previously-failed checks (CI-friendly, exit 0 = all fixed) |
|
|
59
|
+
|
|
60
|
+
### Stack Orchestration
|
|
61
|
+
|
|
62
|
+
| Command | Description |
|
|
63
|
+
|---|---|
|
|
64
|
+
| `stack status` | Show all stack components, install state, config state |
|
|
65
|
+
| `stack plan` | Preview what would be deployed based on current audit (no changes) |
|
|
66
|
+
| `stack deploy` | Deploy stack components (--invariant, --ironcurtain, --all) |
|
|
67
|
+
| `stack sync` | Regenerate stack configs from latest audit — run after harden/fix |
|
|
68
|
+
| `stack teardown` | Remove deployed stack components |
|
|
69
|
+
|
|
70
|
+
### History & Monitoring
|
|
71
|
+
|
|
72
|
+
| Command | Description |
|
|
73
|
+
|---|---|
|
|
40
74
|
| `trend` | ASCII chart of your security score over time |
|
|
41
75
|
| `compare` | Compare coverage vs openclaw security audit |
|
|
42
|
-
| `
|
|
76
|
+
| `log` | View the audit event log |
|
|
77
|
+
| `digest` | Show weekly security digest |
|
|
78
|
+
| `watch` | Monitor config and skill changes in real time |
|
|
79
|
+
| `protect --install` | Install guard hook, shell intercept (zsh/bash/fish), and watch daemon |
|
|
43
80
|
| `snapshot` | Save a config snapshot manually (auto-saved before every harden/fix) |
|
|
44
81
|
| `rollback` | Restore config from auto-snapshot (--list, --id <id>) |
|
|
45
|
-
| `harden --monitor` | Enable monitor mode — observe before enforcing |
|
|
46
|
-
| `harden --monitor-report` | Show what monitor mode has observed |
|
|
47
|
-
| `harden --monitor-off` | Disable monitor mode |
|
|
48
82
|
|
|
49
83
|
## What it catches
|
|
50
84
|
|
|
@@ -59,13 +93,14 @@ clawarmor audit
|
|
|
59
93
|
| Live gateway auth | WebSocket probe — does server actually reject unauthenticated connections? | Full |
|
|
60
94
|
| CORS misconfiguration | OPTIONS probe with arbitrary origin | Full |
|
|
61
95
|
| Gateway exposure | TCP-connects to every non-loopback interface | Full |
|
|
62
|
-
|
|
|
96
|
+
| Multi-step attack chains | read→exfil, inject→execute flows (via Invariant) | Full (with stack) |
|
|
97
|
+
| Runtime tool call interception | Policy-enforced sandboxing (via IronCurtain) | Full (with stack) |
|
|
63
98
|
|
|
64
99
|
## Safety features
|
|
65
100
|
|
|
66
|
-
**Impact classification** — Every fix is tagged 🟢 Safe, 🟡 Caution, or 🔴 Breaking. `--auto`
|
|
101
|
+
**Impact classification** — Every fix is tagged 🟢 Safe, 🟡 Caution, or 🔴 Breaking. `--auto` skips breaking changes unless you pass `--force`.
|
|
67
102
|
|
|
68
|
-
**Config snapshots** —
|
|
103
|
+
**Config snapshots** — Auto-saves before every `harden` or `fix` run:
|
|
69
104
|
|
|
70
105
|
```bash
|
|
71
106
|
clawarmor rollback --list # see all snapshots
|
|
@@ -73,7 +108,7 @@ clawarmor rollback # restore the latest
|
|
|
73
108
|
clawarmor rollback --id <n> # restore a specific one
|
|
74
109
|
```
|
|
75
110
|
|
|
76
|
-
**Monitor mode** — Observe what `harden` would
|
|
111
|
+
**Monitor mode** — Observe what `harden` would change before enforcing:
|
|
77
112
|
|
|
78
113
|
```bash
|
|
79
114
|
clawarmor harden --monitor # start monitoring
|
|
@@ -87,6 +122,8 @@ ClawArmor runs entirely on your machine — no telemetry, no cloud, no accounts.
|
|
|
87
122
|
It has zero npm runtime dependencies, using only Node.js built-ins.
|
|
88
123
|
Every run prints exactly what files it reads and what network calls it makes before executing anything.
|
|
89
124
|
|
|
125
|
+
The full security stack for AI agents doesn't exist as one product. ClawArmor is the foundation that ties it together.
|
|
126
|
+
|
|
90
127
|
## License
|
|
91
128
|
|
|
92
129
|
MIT
|