The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (SAFE): Documentation and CI/CD templates (e.g., references/ci-cd-patterns.md) provide examples for downloading security binaries like Gitleaks and Trivy from GitHub. These sources are included in the Trusted External Sources list, and the patterns represent industry standard deployment practices.
[DATA_EXFILTRATION] (SAFE): The Python scripts epss-integration.py and kev-checker.py perform network requests to api.first.org and cisa.gov. These are reputable security organizations. The data sent is limited to CVE identifiers, and the data retrieved is public vulnerability intelligence. No sensitive local data exposure was found.
[COMMAND_EXECUTION] (SAFE): The skill contains several Bash scripts (basic-scan.sh, vulnerability-report.sh) that invoke trivy, jq, and docker. These executions are consistent with the skill's primary purpose of container scanning and security reporting. Input variables are handled with standard defaults, minimizing injection risks.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill processes external vulnerability data (CVE IDs and descriptions). While this represents a data ingestion surface, the processing is handled by structured scripts that output reports or JSON, rather than being interpolated into an LLM prompt that controls agent logic. Boundary markers are inherently maintained by the structured nature of the tool outputs (SARIF, JSON).

managing-vulnerabilities