setup-harness
Pass
Audited by Gen Agent Trust Hub on Nov 19, 2026
Risk Level: SAFEPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill generates environment setup scripts (
cloud-setup.sh) and instructions that usesudo apt-get installto install system-level dependencies such as the .NET SDK and ShellCheck from official Ubuntu repositories. - [COMMAND_EXECUTION]: The generated
scripts/check.shand.claude/hooks/check-hook.shscripts utilize shell commands to run development tools (e.g.,ruff,eslint,go test) discovered during the repository analysis phase. - [EXTERNAL_DOWNLOADS]: The skill automatically fetches package metadata and version information from well-known public registries including npm, PyPI, Go Proxy, crates.io, NuGet, and Maven Central to ensure tools are pinned to recent, stable versions.
- [REMOTE_CODE_EXECUTION]: The skill downloads and executes binary tools and JAR files from trusted external sources (e.g., Eclipse's repository for
jdtlsand Maven Central forgoogle-java-format). These downloads are protected by SHA-256 integrity checks performed at runtime. - [DYNAMIC_EXECUTION]: The generated harness scripts employ
evalandbash -cto execute development commands constructed at runtime based on the repository's detected structure and configuration files. - [INDIRECT_PROMPT_INJECTION]: The skill's automated discovery process ingests untrusted repository data (manifests like
package.jsonorpyproject.toml) to determine the harness configuration. This surface is mitigated by shell-quoting of filenames and mandatory human-in-the-loop confirmation of all generated configuration diffs.
Audit Metadata