The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages and accesses highly sensitive authentication materials, including private SSH keys (e.g., ~/.ssh/id_ed25519), server login passwords, and sudo passwords for privilege escalation.
[DATA_EXFILTRATION]: Contains instructions to read and display the contents of sensitive configuration files (e.g., cat <DEPLOY_PATH>/.env) from the remote server, which exposes potentially secret environment variables directly to the agent's conversation context.
[COMMAND_EXECUTION]: Facilitates arbitrary command execution on remote infrastructure via SSH. This includes performing administrative tasks using sudo, such as creating system directories and changing file ownership.
[REMOTE_CODE_EXECUTION]: Includes a procedure to download and execute a shell script from Docker's official website (https://get.docker.com | sh) for automated environment setup.
[EXTERNAL_DOWNLOADS]: Fetches configuration and installation resources from well-known external domains associated with Docker and GitHub.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its processing of external, untrusted data that could contain malicious instructions.
Ingestion points: Retrieves and processes remote application logs (docker compose logs) and incoming code/metadata from Git repositories (git pull).
Boundary markers: None identified to separate system instructions from processed external data.
Capability inventory: Extensive capabilities including remote shell access, privilege escalation (sudo), and file transfer (scp).
Sanitization: No evidence of sanitization or filtering for data retrieved from logs or external repositories before it is presented to the agent.

deploy