The Agent Skills Directory

[SAFE]: The skill implements a legitimate multi-turn interaction pattern for software planning. It relies on local reference files (patterns-guide.md, performance-checklist.md) and an output template (architecture-template.md) to generate its final synthesis, avoiding any external dependencies or remote resource fetching.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests user requirements and uses them to populate an architectural plan. Evidence Chain: 1. Ingestion points: Responses to questions in Phase 1 and Phase 2 within SKILL.md. 2. Boundary markers: The skill does not use specific delimiters to isolate user input. 3. Capability inventory: No dangerous tools such as network access, file system writes, or arbitrary command execution are used by the skill. 4. Sanitization: No explicit sanitization or filtering of user input is performed. The absence of dangerous capabilities renders this surface safe.

app-architect