agentic-ecology-bioacoustics
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill's instructions and assets are entirely consistent with its stated purpose of bioacoustic audio analysis. No signs of malicious intent, obfuscation, or unauthorized credential access were found.
- [EXTERNAL_DOWNLOADS]: The skill utilizes official Google Cloud skills and standard high-performance libraries such as TensorFlow and Apache Beam. These are sourced from trusted/well-known directories and are necessary for the bioacoustic analysis tasks and cloud integration.
- [COMMAND_EXECUTION]: Instructions guide the agent in using the gcloud CLI for Google Cloud Storage management and Dataflow job submission. These operations represent standard, authenticated cloud middleware interactions and use project-specific placeholders.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests audio recordings and user-provided annotations, creating an indirect prompt injection surface. The risk is minimized by the skill focusing on numeric embedding processing and category labels. Evidence Chain: (1) Ingestion points: audio files and database annotations (SKILL.md, assets/server.py). (2) Boundary markers: Not explicitly required for the data-heavy task. (3) Capability inventory: gcloud CLI subprocess execution and filesystem writes. (4) Sanitization: assets/server.py implements path sanitization to prevent directory traversal issues in dataset metadata.
Audit Metadata