The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external Java source code, JFR recordings, and JMH benchmark results. This creates a potential surface for indirect prompt injection where malicious instructions could be embedded in code comments or metadata of the files being optimized.
Ingestion points: Processes user-provided Java files (e.g., in OrderIndex.java) and tool outputs like JMH JSON and JFR profiles.
Boundary markers: The skill does not define specific delimiters to separate untrusted data from instructions during the analysis phase.
Capability inventory: The agent can write modified Java files, execute build commands (Maven), and run performance tests (Java, C compiler).
Sanitization: The skill relies on standard tool parsing (JSON, JFR) and does not specify explicit sanitization for instructions embedded in data.
[COMMAND_EXECUTION]: The skill workflow involves executing several shell-based tools to build and profile applications. This includes running Maven (mvn), the Java compiler/runtime (javac, java), and a C compiler (cc) via the assets/examples/verify.sh script.
[DYNAMIC_EXECUTION]: The skill includes examples of JNI and Foreign Function & Memory (FFM) API usage, which involves compiling a C library (native_add.c) at runtime and loading the resulting binary into the JVM for benchmarking. This is a legitimate part of testing native call overhead but involves dynamic code loading.

optimize-java-code