Test fixture
Secure code changes, vulnerability recognition, and safe defaults.
The model receives the prompt (and optional system message). The run uses scorer rubric_json_metrics with the JSON configuration below. Pass/fail and partial credit are determined entirely by that scorer against the model output; no human grading.
Return JSON only with keys verdict, risk, mitigation. A file download endpoint constructs a file path by directly appending a user-provided filename to a base directory without any path normalization or validation.
{
"metrics": {
"correct": {
"checks": [
{
"contains": [
"path traversal"
]
},
{
"contains": [
"../"
]
},
{
"contains": [
"normalize"
]
},
{
"contains": [
"sanitize path"
]
},
{
"contains": [
"restrict directory"
]
}
]
},
"hidden": {
"checks": [
{
"contains": [
"canonical path"
]
},
{
"contains": [
"base directory"
]
},
{
"contains": [
"allowlist"
]
},
{
"contains": [
"chroot"
]
},
{
"contains": [
"realpath"
]
}
]
}
}
}temperature
0
max_tokens
420
timeout (s)
120
type
scored
file
sec-directory-traversal.json