The Agent Skills Directory

[OBFUSCATION]: The skill employs extreme content repetition which can be used to hide malicious intent or exhaust the agent's context window.
Eight files in the references/ directory (including architecture-patterns.md, code-organization.md, and state-management.md) each contain 150 sections of nearly identical technical filler text.
This redundant technical noise serves no functional purpose and is a characteristic pattern for bypassing static analysis filters or burying malicious payloads in a wall of text.
[INDIRECT_PROMPT_INJECTION]: The skill's core workflow for processing architectural reviews creates a significant vulnerability to indirect prompt injection.
Ingestion points: Untrusted user-supplied data enters the agent context via the 'Architecture Review Submission' template defined in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions for the agent to ignore any embedded natural language instructions within the submissions.
Capability inventory: The agent is tasked with creating decision records, assessments, and issues (e.g., via gh issue create as shown in implementation patterns), providing a vector for a successful injection to trigger unauthorized repository actions.
Sanitization: There is no evidence of filtering or validation logic to sanitize external content before the agent processes it for governance decisions.

enterprise-architecture-governance