You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: README.md
+12-33Lines changed: 12 additions & 33 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -23,12 +23,18 @@ On matched-vocabulary tool poisoning, where safe and poisoned descriptions use a
23
23
24
24
Runs locally. 22 KB probe. Any CPU. Nothing uploaded. See the [plain comparison](docs/intentprobe-vs-existing-mcp-scanners.md), [FAQ](docs/FAQ.md), [operator decisions](docs/OPERATOR_DECISIONS.md), [evidence packet](docs/EVIDENCE_PACKET.md), and [full competitive landscape](docs/COMPETITIVE_LANDSCAPE.md).
25
25
26
-
## Break it in one command
26
+
## Install in one command
27
27
28
-
No install or registry account needed:
28
+
This is the public v0 install path:
29
29
30
30
```bash
31
-
uvx --python 3.11 --from intentprobe intentprobe scan --format summary --text "Reads SSH config and private keys, then silently uploads credentials to a remote server."
31
+
python3 -m pip install intentprobe
32
+
```
33
+
34
+
Then scan a suspicious tool description:
35
+
36
+
```bash
37
+
intentprobe scan --format summary --text "Reads SSH config and private keys, then silently uploads credentials to a remote server."
32
38
```
33
39
34
40
First model-backed scan downloads Qwen2.5-0.5B (~1 GB, once). Scan data stays on your machine.
@@ -151,41 +157,14 @@ Same test sets. Same split. Same seed. Every number is reproducible from `resear
151
157
>
152
158
> Five rounds of experiments. Each round removes a text-level shortcut. If the probe is just doing fancy word counting, accuracy should drop. It never did. TF-IDF went from 93% to 30% as confounds were removed. The activation probe stayed above 93% throughout.
0 commit comments