The Agent Skills Directory

[Indirect Prompt Injection] (HIGH): The skill's primary workflow involves processing external user requirements to generate technical designs and a plan.md file, which then triggers automated skill chaining.
Ingestion points: User-provided architectural requirements and existing project documentation.
Boundary markers: Absent. There are no instructions to the agent to treat user requirements as data rather than instructions, nor are there delimiters to wrap external content.
Capability inventory: The skill is granted Read, Write, and Edit permissions, and is explicitly commanded to invoke other skills using the Skill({skill: ..., args: ...}) pattern.
Sanitization: Absent. The arguments passed to subsequent skills (e.g., sw-frontend:frontend-architect) are derived directly from the technology stack defined in the user-influenced plan.
[Unverifiable Dependencies] (MEDIUM): The skill uses a 'phased loading' approach referencing three files in a phases/ directory (01-analysis.md, 02-adr-creation.md, 03-diagrams.md). These files are not included in the provided package, meaning they could contain malicious instructions or persona overrides that remain hidden during static analysis.
[Command Execution] (LOW): The skill is granted Edit and Write tools. While necessary for its stated purpose of creating ADRs and diagrams, these capabilities provide a high-impact vector for data modification if the agent is successfully manipulated via indirect injection.

architect