The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes platform-level tools to manage agent lifecycles and task orchestration. Specifically, it executes TeamCreate, TaskCreate, and Agent commands to spawn a multi-agent hierarchy consisting of a PM, Designer, QA, and Developer agents. While these are internal framework calls, they represent the execution of automated orchestration logic based on user initialization.
[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection by ingesting unsanitized user input (Project name, Business goal, Key features) and writing it directly into persistent storage files like config.json, goals.md, and MEMORY.md. These files serve as the primary context for the spawned sub-agents.
Ingestion points: Step 1 collects free-text project information from the user.
Boundary markers: The generated files do not use explicit delimiters or instructions to ignore embedded commands within the user-provided project data.
Capability inventory: The skill has the capability to write to the filesystem (~/.ai-software-agency/) and spawn multiple sub-agents with specific roles.
Sanitization: There is no evidence of filtering or escaping for the user-provided strings before they are interpolated into the Markdown and JSON templates.

start-project