Security audit
Security checks for vulnerabilities and agentic risk
This skill needs review because it offers vulnerability repair and lasting alert-suppression changes beyond its advertised inspection and alert-status scope.
Prefer the query-only IAM policy scoped to your target project. Enable writes only if you intentionally want repair or whitelist changes, with the exact targets, consequences, and rollback explained before confirmation. Configure keys yourself and review and verify the CLI installer.
ChangeVulStatus explicitly includes immediate_repair, ignore, and add_to_whitelist despite the declared limit of queries plus alert-status marking. These operations can alter host software or suppress vulnerability visibility; confirmation reduces accidental execution but does not resolve the misleading scope.
Changing alert status is consistent with the description, but the command does not constrain operate_type to status-only operations. It therefore exposes response actions that the description explicitly excludes. The supplied chunk does not demonstrate that such an action was executed, but its available capability exceeds the declared scope.
The documented vulnerability-status command permits repair and suppression outside the advertised read-only and alert-marking boundary. An agent following this section could request consequential changes under a workflow the user reasonably understood as inspection-oriented, although explicit confirmation remains required.
Without declared permissions the skill's intent is opaque and cannot be validated.
Adding alarm or login whitelist entries changes future security treatment rather than merely marking an existing alert's status. Presenting these operations as ordinary alert handling can lead to suppression of subsequent detections, even though confirmation is required.
Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.
| ID | Criterion |
|----|-----------|
| NF-01 | All query operations are read-only and safe to execute without confirmation |
| NF-02 | All mutating operations (ChangeEvent, ChangeVulStatus) require explicit user confirmation |
| NF-03 | No AK/SK values are hardcoded in any file |
| NF-04 | Output is structured JSON suitable for programmatic consumption |
The guide downloads a mutable latest shell script and immediately executes it without integrity verification or review. HTTPS protects transport but not against a compromised distribution source; a substituted installer could execute arbitrary code with the invoking user's privileges, potentially exposing cloud credentials.
The diagram routes the handling workflow to ChangeVulStatus while its prose describes handling only as alert-status updates. This introduces a vulnerability-mutation path beyond the declared boundary, although the diagram requires confirmation and does not itself specify a repair or suppression operation.
“Inspection” and “incident” can refer to many activities unrelated to host security. Although the diagram identifies an HSS skill, it does not specify whether these terms require an explicit host-security request, leaving the activation boundary ambiguous.