The Agent Skills Directory

[Privilege Escalation Consideration]: The skill describes a self-hosting method that utilizes the Docker socket for managing sandboxes. As noted in the instructions, accessing the Docker socket is equivalent to root-level permissions on the host system. The skill provides appropriate guidance to use a dedicated host or a managed sandbox service to mitigate this risk.
[Command Execution Pattern]: The deployment process relies on several command-line operations using uv run and docker compose. These are used to build, test, and publish the agent's runtime environment and are standard for software deployment workflows.
[Indirect Prompt Injection Surface]: The skill facilitates the connection of the agent to Slack and external APIs. This creates an ingestion surface where the agent processes external messages. While the skill mentions sandboxing as a mitigation, the ingestion of untrusted data from Slack is a potential vector for indirect instructions.
[Credential Management]: The instructions involve handling sensitive keys such as LANGSMITH_API_KEY, SLACK_BOT_TOKEN, and POSTGRES_PASSWORD. The skill explicitly advises against printing these keys and recommends confirming which variables are forwarded as secrets, which aligns with standard secret management practices.

design-agent-deploy