The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates orchestrator logic for autonomous agents that employs eval() to process state-dependent preconditions. This allows for dynamic decision-making based on variable state inputs during the execution of generated skills.
[INDIRECT_PROMPT_INJECTION]: The skill's design involves reading project-specific architecture documentation and external search results to influence the generation process. This introduces an attack surface where untrusted data could influence the generated skill's configuration or instructions.
Ingestion points: .workflow/codebase/ARCHITECTURE.md (read via Read tool in SKILL.md), and results from maestro search operations.
Capability inventory: Utilizes Bash, Write, Agent, and AskUserQuestion across its phases.
Boundary markers: The skill uses standard markdown headers and code blocks to separate content; however, there are no explicit instructions to ignore potentially adversarial content within ingested files.
Sanitization: The skill does not implement specific filtering or sanitization for project-derived strings before they are interpolated into the generated phase and action templates.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool and the maestro CLI to perform file system operations, load specifications, and delegate complex tasks to other LLM agents. The provided templates also facilitate the creation of custom Bash and Python scripts for deterministic task execution.

skill-generator