The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates executable scripts at runtime. It writes a bash script to /tmp/fetch_sonar.sh and a Node.js script to /tmp/sonar_analyze.js, which are then executed to perform data fetching and parsing. While the code is defined within the skill, runtime generation and execution of scripts is a higher-risk pattern.
[COMMAND_EXECUTION]: The skill invokes several shell commands including curl, chmod, node, cat, and rm to manage its lifecycle and interact with external APIs.
[CREDENTIALS_UNSAFE]: The documentation includes a setup instruction that suggests verifying the $SONARQUBE_TOKEN environment variable by running echo $SONARQUBE_TOKEN. This practice can leak sensitive credentials into shell history files and terminal logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external source, creating an attack surface.
Ingestion points: Data is fetched from the SonarCloud API search endpoint (https://sonarcloud.io/api/issues/search).
Boundary markers: The generated report format lacks delimiters or warnings to the agent to ignore instructions embedded in the issues (e.g., in rule messages or file names).
Capability inventory: The environment allows shell command execution, file system operations in /tmp, and network access via curl.
Sanitization: The Node.js analysis script extracts fields like message and rule directly from the API response and includes them in the output without sanitization or escaping.
[DATA_EXFILTRATION]: The skill transmits the SonarCloud token via curl to https://sonarcloud.io. This is documented as a neutral finding as it communicates with a well-known service for its primary intended purpose.

sonarqube