build-project
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes build commands and local scripts found within the target project root, such as
./gradlew buildandmvn package. This involves running project-local scripts that are part of the external content being analyzed, which is a standard part of project builds but carries inherent risk if the project is untrusted. - [EXTERNAL_DOWNLOADS]: The workflow requires resolving project dependencies via build tools and updating submodules using
git submodule update --init. These operations pull content from external package registries and Git repositories into the environment. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes untrusted project files (e.g.,
pom.xml,build.gradle, source code) to determine build parameters, modules, and toolchains. Maliciously crafted content in these files could attempt to manipulate the agent's behavior during the detection or configuration phases. - Ingestion points: Project configuration files (
pom.xml,build.gradle,settings.gradle) and source code within the user-provided<project-root>. - Boundary markers: Absent. The skill does not instruct the agent to use delimiters or ignore instructions within the project files it reads.
- Capability inventory: Execution of build tools (
gradlew,mvn), modification of project configuration files, and file system deletion (.opentaint/project). - Sanitization: Absent. Data read from the project root is used to determine workflow steps and build commands without explicit validation or escaping of the content.
Audit Metadata