The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions do not contain any patterns designed to bypass agent safety filters, override system instructions, or extract system prompts.
[CREDENTIALS_UNSAFE]: No hardcoded API keys or tokens are present. The documentation correctly identifies the need for environment variables (SONARQUBE_TOKEN) and provides secure setup instructions.
[EXTERNAL_DOWNLOADS]: The skill references the official SonarSource GitHub repository for the MCP server and its official Docker Hub image. These are well-known and trusted sources for the integrated service.
[DATA_EXFILTRATION]: No patterns of unauthorized data harvesting or transmission to non-whitelisted domains were identified. Tool outputs are directed back to the agent for user presentation.
[INDIRECT_PROMPT_INJECTION]: The skill includes functionality to ingest and analyze external code snippets, which represents a potential surface for indirect injection.
Ingestion points: The analyze_code_snippet tool (documented in SKILL.md and references/llm-context.md) accepts a fileContent parameter containing untrusted code.
Boundary markers: Not explicitly implemented in the instruction set for tool parameters.
Capability inventory: The agent has the ability to view project status, search issues, and update issue states using change_sonar_issue_status.
Sanitization: No explicit sanitization steps for the input code are detailed, but the skill mandates user confirmation for status changes.
[COMMAND_EXECUTION]: The skill operates entirely through structured MCP tools and does not attempt to execute arbitrary shell commands or manipulate the host filesystem beyond the allowed scope.

sonarqube-mcp