The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill repeatedly instructs developers to exclude credentials, auth material, and private records from source control, build bundles, and simulator sessions. It explicitly warns against hardcoding secrets or asking users to paste them into chat.
[COMMAND_EXECUTION]: The instructions utilize standard development tools including bun, bunx, and cargo for scaffolding and building projects. It promotes safer dependency installation by recommending the use of --frozen-lockfile and --ignore-scripts to prevent the execution of potentially malicious lifecycle scripts.
[EXTERNAL_DOWNLOADS]: The skill documents the platform's ability to install plugins directly from GitHub repositories. It provides clear guidance for users to review the trust and requested permissions of the source repository before proceeding with installation.
[INDIRECT_PROMPT_INJECTION]: The guidance addresses potential injection vectors from external data sources such as deployment events and UI message bridges. It instructs developers to validate origins, verify data fields, and defend against SSRF and DNS rebinding in networked plugins.
[PRIVILEGE_ESCALATION]: The documentation clarifies the security model by stating that native plugins run with the host OS account's permissions and advising against describing the plugin environment as sandboxed to avoid misleading users about the actual level of isolation.

temps-plugin