The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The documentation in SKILL.md instructs the user to install the platform using a remote script piped directly into a shell (curl -sSL https://dokploy.com/install.sh | sh). This is a dangerous execution pattern that allows arbitrary code from a remote server to run on the host system without prior verification.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary commands on the host and within containers. Specifically, platform-ops-security.md describes using docker exec for scheduled application jobs and running bash scripts directly on remote servers for maintenance tasks. These functions provide the agent with high-privilege primitives for system manipulation.
[PROMPT_INJECTION]: An indirect prompt injection surface is defined in deployment-and-builds.md through the 'Preview Deployments' feature.
Ingestion points: Untrusted data enters the agent's deployment environment via GitHub Pull Requests (deployment-and-builds.md).
Boundary markers: The skill provides a explicit warning: 'Do not enable for public repos — external users could run builds on your server.' It also suggests a PR label filter to restrict triggers.
Capability inventory: The environment includes image builders (Nixpacks, Dockerfile), container management tools (docker exec), and persistence mechanisms via cron jobs (platform-ops-security.md).
Sanitization: No technical sanitization or content validation for the PR data is mentioned, relying solely on user-configured filters.

dokploy-best-practices