knowledge-catalog
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external repository content and interacts with Google Cloud resources, which could contain malicious instructions.
- Ingestion points: Files within the GoogleCloudPlatform/knowledge-catalog repository and metadata from Google Cloud Dataplex/Catalog.
- Boundary markers: The instructions explicitly require user approval for all external effects and warn that local validation does not prove cloud state.
- Capability inventory: The skill uses Bash, Read, Grep, and Glob tools and is instructed to perform cloud writes and resource creation via kcmd and reference-agent.
- Sanitization: The preflight.py script validates repository file hashes and sanitizes file paths to prevent traversal and access to sensitive folders like .ssh or .aws.
- [EXTERNAL_DOWNLOADS]: The skill references and is designed to interact with a specific revision of the GoogleCloudPlatform/knowledge-catalog repository on GitHub.
- Evidence: SKILL.md and source.json pin the repository to commit 6888b305651f366d878b084f68f86ff2c3effe2e.
- [COMMAND_EXECUTION]: The skill instructions involve executing CLI tools like kcmd, reference-agent, and python3.
- Evidence: SKILL.md and references/upstream.md detail the use of these tools for semantic model emission, deployment, and preflight checks. Access and execution are restricted by explicit approval gates defined in the instructions.
Audit Metadata