The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes project source code and user-provided scope strings to generate unit, integration, and E2E tests. This data is passed into sub-agent prompts (ork:test-generator and ork:emulate-engineer) that have permissions to execute shell commands and modify the filesystem. The skill lacks explicit delimiters or instructions to prevent sub-agents from following malicious commands that might be embedded in code comments, metadata, or documentation within the analyzed project.
Ingestion points: User-provided SCOPE arguments and project source files (e.g., src/**/*.ts, src/**/*.js, test_*.py).
Boundary markers: Absent in the prompts dispatched to sub-agents.
Capability inventory: The skill and its sub-agents use Bash, Write, Edit, Agent, and Workflow tools.
Sanitization: No validation or sanitization is performed on external code before it is interpolated into prompts.
[COMMAND_EXECUTION]: The skill automatically executes development tools and test runners (e.g., npm test, pytest, npx vitest) and infrastructure commands (e.g., docker compose). While essential for the skill's purpose, these operations rely on the integrity of the project's local configuration files (like package.json or docker-compose.yml), which could be leveraged to execute arbitrary code if those files are malicious.

cover