The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the endorctl command-line tool to perform lookups. These operations are strictly limited to read-only API calls (such as list) to fetch package risk metadata, with a prohibition on mutating commands like scan, update, or delete.
[DATA_EXFILTRATION]: Accesses the local configuration file at ~/.endorctl/config.yaml to identify the active namespace. The skill includes explicit security instructions to never disclose or echo credentials, secrets, tokens, or the full contents of the configuration file.
[PROMPT_INJECTION]: The skill processes potentially untrusted external data, including repository comments and package metadata. To mitigate the risk of indirect prompt injection, the agent is explicitly instructed to treat all such external content as data rather than instructions.

package-risk-summary