ps-architect
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill defines a structured methodology for software architecture, involving phases for grounding, sketching, and implementation. Analysis of the instructions and reference templates revealed no evidence of malicious intent, credential theft, or unauthorized persistence mechanisms. The tool utilizes local configuration for model selection and focuses on generating design artifacts.
- [INDIRECT_PROMPT_INJECTION]: The design process ingests grounding artifacts which are used to generate implementation sketches. This workflow involves a data ingestion surface where external content could influence agent output.
- Ingestion points: Grounding artifacts and task descriptions processed during the 'Sketch' phase in SKILL.md and references/runner-prompt.md.
- Boundary markers: The skill uses specific templates (rationale-template.md) and red-flag screening (design-red-flags.md) to structure the output, though it lacks explicit delimiters for untrusted data.
- Capability inventory: Synthesis of design rationales and implementation of code against sketches (Phase D), involving filesystem writes and calls to peer skills like ps-arena.
- Sanitization: No explicit sanitization or escaping of grounding artifacts is mentioned before they are passed to the runner prompt.
Audit Metadata