The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied design concepts by passing them through a series of specialized sub-agents (e.g., draft, critic, judge, and redesign agents). Malicious input could attempt to influence the behavior of these agents.
Ingestion points: The concept argument provided by the user is passed to multiple agents across the workflow files.
Boundary markers: The skill employs clear markers such as --- SPEC START --- and CORE_MECHANISM_START to delimit content.
Capability inventory: The skill uses spawn_subagent to execute parallel tasks and can deliver files to external Slack endpoints.
Sanitization: Input is passed via file paths to sub-agents without explicit sanitization or filtering logic.
[COMMAND_EXECUTION]: The execution model depends on the sagaflow CLI to launch the durable workflow and manage background tasks.
[DATA_EXFILTRATION]: The skill delivers completed design specifications and progress reports to an external Slack workspace. This is a standard feature for the author's ecosystem, utilizing Slack as a well-known service for artifact delivery.

deep-design