The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions to read codebase dependency files such as requirements.txt and pyproject.toml to determine the project variant. This creates a surface for indirect prompt injection.
Ingestion points: Project dependency declarations and imports within the user's codebase.
Boundary markers: None defined; the agent reads the file content directly.
Capability inventory: The agent's capabilities in this context are limited to generating text-based architectural designs, diagrams, and review reports.
Sanitization: No specific sanitization or filtering of the input file content is implemented.
[DATA_EXFILTRATION]: The skill is designed to identify specific library imports (e.g., nf_common, bclearer_pdk) within the project environment. This metadata access is necessary for the skill's primary function and does not involve accessing sensitive files, credentials, or performing unauthorized network transmissions.
[SAFE]: The skill uses clear, non-obfuscated language and does not contain any hardcoded credentials, remote code execution triggers, or persistence mechanisms. The internal "severity levels" mentioned in the review mode are strictly for architectural assessment and do not interfere with the agent's safety protocols.

ob-architect