The Agent Skills Directory

[SAFE]: The skill implements a multi-category 150-point static analysis pipeline using local Python scripts (validate_apex.py, llm_pattern_validator.py). These scripts use regular expressions and structural analysis to detect critical anti-patterns like SOQL/DML in loops, security vulnerabilities (sharing keyword violations, SOQL injection), and LLM-specific hallucinations.
[SAFE]: All Salesforce metadata operations are performed through the Model Context Protocol (MCP) using established tooling API and metadata API conventions. The skill includes mandatory guardrails that warn or block the agent if dangerous patterns are detected in the code before deployment.
[SAFE]: File system access is restricted to the skill's own directories for script execution and the system temporary directory for transient validation files, following standard practices for local tooling.
[SAFE]: No evidence of prompt injection, data exfiltration, obfuscation, or unauthorized privilege escalation was found in the skill's instructions or supporting scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests Apex code from external sources (Salesforce org or user input) for validation. While this presents an attack surface where malicious instructions could be embedded in the code comments or strings, the risk is minimized by the use of static analysis scripts that operate independently of the LLM's direct prompt processing, and the skill's instructions emphasize using these script results as the primary source of truth. (Severity: SAFE)

sf-apex