The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references a documentation URL https://jeffallan.github.io/claude-skills/skills/language/java-architect/ that is blacklisted by automated reputation scanners (URLite).
[REMOTE_CODE_EXECUTION]: The skill directs the agent to use Maven and Gradle wrappers (./mvnw verify and ./gradlew check). These tools download and execute external code dependencies and run tests, which can be exploited to achieve remote code execution in the agent's environment.
[COMMAND_EXECUTION]: The Core Workflow in SKILL.md requires the execution of shell commands such as ./mvnw verify -pl <module> and ./gradlew check. Executing these scripts on untrusted or attacker-controlled projects provides a direct path for shell-based attacks.
[INDIRECT_PROMPT_INJECTION]: The skill is highly vulnerable to indirect prompt injection because it processes Java projects and executes commands based on their content without sanitization or boundary markers.
Ingestion points: Java source code, pom.xml files, build.gradle scripts, and integration test suites.
Boundary markers: None present; the skill lacks instructions to isolate or verify the project files before execution.
Capability inventory: Execution of build tools (Maven/Gradle) with network access and file system write permissions.
Sanitization: None; the agent is instructed to run tests and resolve failures based on the output of the potentially malicious code.
[METADATA_POISONING]: There is a discrepancy between the provided vendor context ('sodiqabdulwaris') and the author metadata ('https://github.com/Jeffallan') in the SKILL.md file, which may be intended to mislead the agent or users regarding the skill's origin.

java-architect