The Agent Skills Directory

[Command Execution] (SAFE): The skill utilizes the Bash tool for standard development tasks such as compilation (mvn compile) and testing (mvn test). These commands are scoped to the project directory and do not exhibit high-risk behavior like privilege escalation or service modification.
[Indirect Prompt Injection] (LOW): The skill processes external source code through tools like Read and search_for_pattern.
Ingestion points: Reads local Java source files via Read and mcp__serena__search_for_pattern.
Boundary markers: Absent; there are no specific delimiters defined to separate code data from instructions.
Capability inventory: Includes Bash for command execution and mcp__serena__replace_symbol_body for file writes.
Sanitization: No explicit sanitization of ingested code content is implemented.
[Data Exposure & Exfiltration] (SAFE): Analysis of the instructions and code snippets shows no access to sensitive paths (e.g., ~/.ssh, ~/.env) or network calls to external domains.
[Obfuscation] (SAFE): No evidence of encoded payloads (Base64), zero-width characters, or homoglyph-based evasion was found.
[Prompt Injection] (SAFE): The skill contains no instructions designed to bypass agent constraints or safety filters.

compiler-dev