The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements 'Context Stuffing' (or 'Context Inflation') by including eight reference files (e.g., architecture-patterns.md, code-organization.md, deployment-pipelines.md, error-handling.md, performance-optimization.md, security-best-practices.md, state-management.md, and testing-strategies.md) that each contain 150 repetitive sections of technical jargon. This excessive data volume (estimated at over 1.5 million words across the files) is a known method for displacing system-level prompts and safety constraints within the AI's context window.
[INDIRECT_PROMPT_INJECTION]: The presence of massive amounts of repetitive technical data in the reference files creates an exploitable attack surface.
Ingestion points: Multiple voluminous technical reference documents in the references/ directory.
Boundary markers: Absent; the repetitive technical blocks are not delimited or marked as untrusted data.
Capability inventory: The skill provides implementation patterns for signaling and media servers which involve network and file system access.
Sanitization: Absent; no validation or escaping logic is provided to handle the voluminous технический content.
[DYNAMIC_EXECUTION]: In references/live-streaming.md, there are PowerShell script artifacts such as '@ | Set-Content -Path D:...' that refer to absolute local file system paths from the author's environment. These could be misinterpreted by an AI agent as commands to generate or modify files on the host system.
[OBFUSCATION]: The extreme redundancy across the reference files serves as a form of noise-based obfuscation, making it difficult to detect hidden malicious instructions within the data and complicating automated security analysis.

backend-web-real-time