The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to create directory structures (e.g., mkdir -p) for the generated skills based on user-influenced path variables.
[REMOTE_CODE_EXECUTION]: The skill generates new agent instructions (role.md files) and logic, then uses the Agent tool to spawn sub-agents that execute this dynamically created content. This dynamic generation and execution of agent instructions is the core function of the skill.
[PROMPT_INJECTION]: The skill incorporates user-provided requirements (domain descriptions, role names, and pipelines) into the instructions of the generated coordinator and worker agents, representing a surface for indirect prompt injection.
Ingestion points: User input for skill name and domain description parsed in Phase 1 (01-requirements-analysis.md).
Boundary markers: No explicit delimiters or sanitization identified for the user-provided domain text within the generated role files.
Capability inventory: Bash (mkdir), Write (file creation), and Agent (spawning sub-agents) tools are available to the generated team.
Sanitization: Input validation for skill and role names (e.g., kebab-case enforcement) is included to mitigate risks such as path traversal in filenames.

team-designer