reproducible-native-bundling-two-tasks

Pass

Audited by Gen Agent Trust Hub on Aug 30, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructions utilize shell command execution through a runChecked wrapper in Gradle to perform archive operations like tar -czf. It also suggests manual verification using shell tools such as shasum, find, and grep.
  • [EXTERNAL_DOWNLOADS]: The nativesSetupAll task is designed to fetch native library tarballs from remote repositories using curl. This is a standard procedure for build systems to manage large binary dependencies.
  • [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from external sources.
  • Ingestion points: External tarball artifacts downloaded via curl from remote servers.
  • Boundary markers: The pattern uses a hardcoded nativesChecksums map in the build.gradle.kts file as a security boundary.
  • Capability inventory: File system writes, subprocess execution (tar), file deletion (archive.delete()), and runtime loading of native libraries.
  • Sanitization: The skill explicitly mandates integrity verification by comparing the SHA-256 digest of the downloaded artifact against the pinned map, failing the build and deleting the artifact if a mismatch is detected.
  • [DYNAMIC_EXECUTION]: The workflow involves unpacking binary artifacts into the application's runtime environment, which are then loaded as native shared objects (.so, .dll, .dylib) by the JVM.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 30, 2026, 07:03 AM
Security Audit — agent-trust-hub — reproducible-native-bundling-two-tasks