The Agent Skills Directory

[SAFE]: The skill provides a framework for architectural review and design, managing role-based orchestration and documentation within the project's agent context. It correctly scopes its operations to project files and does not exhibit malicious intent.
[INDIRECT_PROMPT_INJECTION]: The skill has an indirect injection surface as it processes external documentation and codebase content. (1) Ingestion points: Reads codebase structure, CONTEXT.md, existing ADRs, and vendor documentation from the knowledge vault. (2) Boundary markers: None explicitly mentioned in the instructions to the agent for separating data from instructions. (3) Capability inventory: Writing session hub files to .agent/context/ and orchestrating model-invoked skills like domain-modeling and research-from-vault. (4) Sanitization: None identified. Severity is assessed as safe given it is the core functionality and no exploitable chains are present.
[DATA_EXFILTRATION]: The skill mentions 'secrets' and 'keys' within the security and infrastructure roles. This is scoped to architectural analysis and design documentation, with no evidence of commands that harvest or exfiltrate actual credential values from the user's environment.

system-architect