The Agent Skills Directory

[CREDENTIALS_UNSAFE]: Hardcoded default credentials are used throughout the skill's documentation and scripts.
Evidence: SKILL.md contains the literal credential string 'admin:Harbor12345' in multiple curl command examples used to demonstrate API interactions.
Evidence: scripts/agent.py includes 'password = "Harbor12345"' as a hardcoded default value within the main function.
[INDIRECT_PROMPT_INJECTION]: The audit scripts are vulnerable to indirect prompt injection by processing untrusted data from registry APIs without proper sanitization.
Ingestion points: scripts/agent.py and scripts/process.py retrieve project metadata, member information, and system audit logs from the Harbor REST API endpoints.
Boundary markers: There are no delimiters or instructions to ignore potential malicious content within the processed JSON data.
Capability inventory: The scripts possess the capability to write to the file system ('harbor_audit_report.json') and output data to the console based on values retrieved from the external API.
Sanitization: No sanitization, escaping, or schema validation is performed on the strings retrieved from the API before they are rendered in reports.
[COMMAND_EXECUTION]: The skill provides an auditing script that implements insecure network communication settings.
Evidence: In scripts/process.py, the harbor_api_call function explicitly disables SSL verification by setting 'ctx.verify_mode = ssl.CERT_NONE'. This insecure configuration allows for Man-in-the-Middle (MitM) attacks, which could lead to the theft of registry credentials or the manipulation of security audit results.

securing-container-registry-with-harbor