The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for C++ networking servers and clients which ingest untrusted data from network sockets, creating a potential surface for indirect injection if that data is later processed by an LLM.
Ingestion points: Socket read operations such as socket_.async_read_some, async_read, and async_read_until are documented in references/coroutines.md and references/pre-cpp20.md.
Boundary markers: The skill explicitly instructs users to implement framing logic (Step 3 in SKILL.md) using length-prefixed binary framing or delimiter-based reads to properly delineate untrusted external inputs.
Capability inventory: The templates enable network communication and data processing; the skill scripts do not include subprocess spawning or arbitrary command execution.
Sanitization: The instructions mandate strict error handling for every asynchronous operation and highlight buffer lifetime management to prevent memory safety issues common in C++ development.
[EXTERNAL_DOWNLOADS]: The skill includes reference links to official documentation for Boost and Asio, as well as worked examples on a GitHub repository (github.com/alexprivalov/boost-asio-skill).
These links are provided for developer reference and documentation purposes. The skill does not contain instructions to automatically download or execute remote code during operation.
[SAFE]: The skill adheres to industry best practices for C++ systems programming, focusing on thread safety, correct use of strands, and proper lifecycle management. No malicious obfuscation, persistence mechanisms, or privilege escalation attempts were detected.

boost-asio-pro