The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted input data (such as logs, traces, database dumps, and conversation history) to construct engineered prompt payloads. This process creates a surface for indirect prompt injection (Category 8) where malicious instructions embedded in the processed data could influence the agent's behavior. While the instructions recommend sanitization and structural delimiters, the provided implementation logic for stripping injections is not fully realized in the code snippets.
[PROMPT_INJECTION]: Multiple reference files, including architecture-patterns.md, security-best-practices.md, state-management.md, testing-strategies.md, and performance-optimization.md, contain 150 sections of identical boilerplate text. This extreme documentation bloating is a suspicious pattern used to obscure the signal-to-noise ratio, potentially bypassing security auditing tools or token-based filtering systems by burying content in redundant noise.

context-engineering