The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script scripts/agent.py executes security scanning tools (Semgrep, Trivy, Gitleaks) using subprocess.run. The implementation follows security best practices by passing commands as lists and avoiding shell invocation, which prevents command injection vulnerabilities from untrusted user inputs provided to the target argument.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it ingests and processes findings from external codebases scanned by security tools. However, the risk is negligible as the data is handled as structured JSON and used for reporting rather than being fed back into an LLM's primary instruction context.
Ingestion points: agent.py parses JSON output from semgrep, trivy, and gitleaks via json.loads.
Boundary markers: None; the script extracts specific fields like check_id, severity, and message from the tool outputs.
Capability inventory: The script is limited to executing predefined scanning commands, writing JSON reports to local files, and printing status messages to the console.
Sanitization: Findings are treated as data points for aggregation and threshold enforcement. The script does not evaluate the content of the findings as code or instructions.

implementing-devsecops-security-scanning