The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains hardcoded credentials in its examples.
Evidence: The credential 'admin:Harbor12345' is used in several curl command examples throughout the SKILL.md file.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using curl and jq to interact with the Harbor API.
Evidence: Multiple examples in SKILL.md demonstrate the use of curl -X POST, curl -s, and pipe operations with jq.
[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch data from an external registry service.
Evidence: Examples show curl commands fetching tokens and repository information from harbor.example.com.
[DATA_EXFILTRATION]: The skill performs network operations to an external domain that is not included in the standard whitelist.
Evidence: curl commands in SKILL.md target https://harbor.example.com.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process data from an external API without defined safety boundaries.
Ingestion points: API responses from Harbor endpoints (such as /projects, /repositories, and /artifacts) enter the agent's context through the examples provided.
Boundary markers: There are no explicit markers or instructions in SKILL.md to differentiate between the skill's trusted instructions and the untrusted data returned by the API.
Capability inventory: The skill utilizes powerful tools like curl and jq via shell execution, which could be misused if an attacker-controlled Harbor instance returns malicious content.
Sanitization: The instructions do not specify any validation, escaping, or sanitization procedures for the data retrieved from the Harbor API before it is handled by the agent.

harbor-api