workflow-integration-test
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes a local Node.js script located at
.claude/scripts/lib/workflow-skill-contract.cjsto update workflow definitions within the project environment. - [DYNAMIC_EXECUTION]: The skill performs dynamic loading and execution of other skills from the
.agents/skills/directory based on a registry and includes functionality to 'provision' Python runtimes on the host system to ensure test readiness. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and external configuration files to orchestrate actions with significant capabilities, creating a potential surface for indirect injection.
- Ingestion points: User prompts, the
.claude/workflows.jsonregistry, and multiple protocol files in.claude/skills/shared/protocols/. - Boundary markers: The skill uses blocking markers for workflow state but lacks specific delimiters or 'ignore' instructions for untrusted data processed during the workflow.
- Capability inventory: The skill can spawn subagents (
spawn_agent), execute local scripts, and write files to thetmp/reports/directory. - Sanitization: There is no documented sanitization or escaping of data before it is interpolated into sub-skill calls or processed by the runner.
Audit Metadata