The Agent Skills Directory

[SAFE]: The skill instructions were found to be secure, focusing on analytical project management tasks. It uses standard tools (Read, Glob, Grep, Write, Task) to process local documentation without any external network access or unauthorized file system operations.- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by consuming user-provided documentation (GDDs, ADRs, stories, and tests) to build its traceability matrix. This is a standard functional requirement for an architecture review tool. The potential for indirect prompt injection is mitigated by the structured nature of the analysis and the fact that extracted requirements are reported rather than executed. * Ingestion points: Markdown files in design/gdd/, docs/architecture/, production/epics/, and tests/. * Boundary markers: The skill uses phase-based processing but lacks explicit markers to disregard instructions within input files. * Capability inventory: Includes Write for file updates and Task for spawning sub-agents. * Sanitization: No explicit sanitization of extracted requirement text is defined.

architecture-review