The Agent Skills Directory

[PROMPT_INJECTION]: A static analysis hint flagged text within the 'Prompt Injection Safety Notice' section. Analysis confirms that the string 'ignore previous instructions' is used purely defensively to instruct the agent to disregard malicious commands that might be present in the user-provided design documents being analyzed.
[SAFE]: The skill follows a least-privilege model, restricting tool usage to read-only file system operations ('Read', 'Grep', 'Glob'). It lacks network access, file write permissions, and dynamic execution capabilities, which effectively prevents data exfiltration and unauthorized system modification.
[SAFE]: The skill correctly handles the ingestion of untrusted architecture documents by establishing clear trust boundaries in its instructions, advising the agent to redact sensitive credentials and maintain role boundaries throughout the analysis process.

threat-modeling