dotnet-nuget-update
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from the target repository by extracting XML comments from project files.
- Ingestion points: XML comments are extracted using the
Get-AdjacentXmlCommenthelper inscripts/_common.ps1and included in the audit results processed byscripts/Get-DependencyAudit.ps1andscripts/Get-PackageGraph.ps1. - Boundary markers: Absent. The instructions in
SKILL.mdexplicitly advise the agent to "Treat comments as maintainer intent" and to "Read the note before applying the update," which could allow a malicious comment to influence agent behavior. - Capability inventory: The skill possesses the ability to perform network requests to NuGet registries and perform file writes to modify project dependencies.
- Sanitization: The scripts do not sanitize the extracted comment text before displaying it to the agent.
- [EXTERNAL_DOWNLOADS]: The skill fetches package metadata from official NuGet registries.
- Evidence:
scripts/Get-DependencyAudit.ps1andscripts/Resolve-NuGetVersion.ps1perform network lookups againsthttps://api.nuget.org/v3-flatcontaineror other user-configured feeds. - Context: These operations are a core part of the skill's intended functionality and target well-known package registry services.
- [CREDENTIALS_UNSAFE]: The skill reads local NuGet configuration files that often contain sensitive credentials.
- Evidence:
scripts/Get-NuGetSources.ps1accessesNuGet.Configfrom standard system locations such as%APPDATA%\NuGet\NuGet.Config(Windows) or~/.nuget/NuGet/NuGet.Config(Linux/macOS). - Context: The tool identifies configured package sources as part of its dependency audit. It includes logic to identify the presence of credentials while specifically excluding the actual password values from its output objects.
Audit Metadata