The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes an evaluation framework that ingests and processes untrusted C++ source files (e.g., evals/files/queued-labels.cpp, evals/files/publication.cpp). This represents a potential attack surface where malicious code comments or identifiers could attempt to influence the agent's behavior.
Ingestion points: C++ source files referenced in evals/evals.json.
Boundary markers: The skill does not define explicit delimiters or instructions for the agent to ignore embedded commands within the code it analyzes.
Capability inventory: The skill has the capability to read files and provide build/diagnostic command templates (e.g., clang++, ctest) to the agent.
Sanitization: No explicit sanitization of ingested code content is described within the skill body.
[EXTERNAL_DOWNLOADS]: The skill references technical documentation, standard library specifications, and source repositories from trusted organizations and well-known services.
Fetches guidance from official Clang/LLVM AddressSanitizer and ThreadSanitizer documentation.
References official CMake build system and feature documentation.
Cites technical references from the isocpp organization and C++ working drafts on GitHub.
[SAFE]: The technical instructions provided in the core rules and workflows are consistent with modern C++ best practices for memory safety and concurrency. No patterns of data exfiltration, credential theft, or persistence were detected.

cpp