tilebox-workflow-releases

Pass

Audited by Gen Agent Trust Hub on Aug 22, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill relies on the tilebox CLI to perform core workflow management tasks. This includes tilebox workflow init for project scaffolding, tilebox workflow publish-release for artifact creation, and tilebox workflow deploy-release for cluster management. These operations are essential to the skill's purpose and are executed within the user's project environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines an 'Implicit Project Route' where the agent interprets geospatial prompts as authorization to initialize and deploy workflows. This presents an attack surface where processed data could influence agent actions.
  • Ingestion points: Implicit geospatial prompts routed to the agent in SKILL.md.
  • Boundary markers: The instructions include explicit constraints, such as requiring 'explicit production intent' and respecting 'code only' or 'do not deploy' directives.
  • Capability inventory: Access to tilebox CLI commands for project initialization, file creation (cat > tilebox.workflow.toml), release publishing, and cluster deployment.
  • Sanitization: The skill emphasizes the use of uv.lock for deterministic dependency management and SHA-256 digests for release artifact verification.
  • [EXTERNAL_DOWNLOADS]: The skill provides examples for configuring package sources in pyproject.toml, specifically referencing the official PyTorch CPU wheel index (https://download.pytorch.org/whl/cpu). These references target well-known, established services for software distribution.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 22, 2026, 08:44 PM
Security Audit — agent-trust-hub — tilebox-workflow-releases