The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill references setup scripts for Grype and Syft from Anchore's official GitHub repositories.
Evidence: The commands curl -sSfL https://raw.githubusercontent.com/anchore/grype/main/install.sh | sh and curl -sSfL https://raw.githubusercontent.com/anchore/syft/main/install.sh | sh are included in the documentation and agent script error messages.
Security Note: These sources represent well-known security tools, and the download patterns are consistent with their official installation procedures.
[COMMAND_EXECUTION]: The scripts/agent.py script invokes security tools such as Trivy, Syft, and Cosign using the subprocess module.
Evidence: The code in scan_image_trivy, generate_sbom, and verify_image_signature uses subprocess.run() with arguments passed as a list.
Security Note: By passing command arguments as a list and avoiding shell=True, the script correctly mitigates the risk of shell command injection from potentially untrusted image references.
[METADATA_POISONING]: There is a minor inconsistency between the author field in the skill's frontmatter ('mahipal') and the copyright owner in the license file ('mukul975'). This is a documentation discrepancy rather than a malicious pattern.
[INDIRECT_PROMPT_INJECTION]: The skill processes container image references, which are external strings that could contain malicious content designed to influence downstream logic.
Ingestion points: Image references provided via the --image flag to scripts/agent.py.
Boundary markers: The script does not use specific delimiters or warnings for image strings, but treats them as literal arguments.
Capability inventory: The agent has permissions to execute binaries, write scan reports to the local file system, and interact with the AWS ECR service.
Sanitization: Input is treated as a literal argument string in subprocess calls, ensuring that shell-specific characters are not interpreted as commands.

securing-container-registry-images