The Agent Skills Directory

[COMMAND_EXECUTION]: The implementation instructions direct the agent to use $ARGUMENTS (user input) within shell command templates. Specifically, the patterns in Step 2 (e.g., grep -q "com/example/ClassName.class") and the use of find and jar commands do not include explicit sanitization or quoting of the class name variable. This could allow for command injection if a user provides a class name containing shell metacharacters like ;, &, or backticks.
[EXTERNAL_DOWNLOADS]: The skill utilizes jbang to retrieve and execute the Vineflower decompiler (org.vineflower:vineflower:RELEASE). This process fetches code from Maven Central, a well-known repository for Java artifacts, to facilitate the decompilation process.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill reads and displays source code decompiled from project dependencies (SKILL.md).
Boundary markers: The instructions lack specific guidance for the agent to treat the decompiled output as untrusted data or to ignore any instructions found within the code comments or logic.
Capability inventory: The skill has the ability to execute shell commands via mvn, jbang, find, jar, and rm (SKILL.md).
Sanitization: There is no evidence of filtering or escaping the decompiled source code before it is presented to the agent's context.

java-decompile