The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to utilize high-privilege shell interfaces including bash, SSH, and docker exec for host-level and container-level operations. Evidence is located in SKILL.md, references/ssh-docker.md, and references/troubleshooting.md.
[COMMAND_EXECUTION]: Guidance in references/source-code-analysis.md explicitly recommends using node -e to dynamically execute JavaScript code for reading files inside running containers, which represents a dynamic execution risk.
[DATA_EXFILTRATION]: The skill outlines procedures for accessing sensitive source code, configuration files, and database schemas using commands such as grep, fs.readFileSync, and psql. While intended for legitimate administration, this provides a mechanism for exposing system secrets. Evidence in references/source-code-analysis.md.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its ingestion of untrusted external data.
Ingestion points: Application logs, container logs (docker logs), and terminal outputs as described in references/ssh-docker.md.
Boundary markers: Absent; there are no instructions to treat log content as data or to ignore embedded instructions.
Capability inventory: Host-level command execution via SSH, container manipulation via Docker, and full resource management via Dokploy MCP tools (references/mcp.md).
Sanitization: Absent; no sanitization or filtering of log data is mentioned before processing.

dokploy-admin