The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents the osmedeus worker eval command, which allows for the evaluation of arbitrary scripts on distributed worker nodes. \n- [REMOTE_CODE_EXECUTION]: The remote-bash step type and ssh_exec utility function are documented for executing shell commands on remote infrastructure via SSH or Docker containers. \n- [COMMAND_EXECUTION]: Extensive documentation is provided for local command execution through standard bash steps, the execCmd parameter generator, and language-specific execution functions like exec_python and exec_ts (using the Bun runtime). \n- [CREDENTIALS_UNSAFE]: Examples and references within the documentation point to sensitive file paths, such as ~/.ssh/scanner_key in the remote-bash configuration and the retrieval of sensitive environment variables using the getEnvVar generator. \n- [COMMAND_EXECUTION]: The skill documents the sudo_auth function and the --sudo-aware flag, which are designed to facilitate the acquisition and maintenance of root privileges during workflow execution. \n- [PROMPT_INJECTION]: The agent step type presents a surface for indirect prompt injection where an LLM processes untrusted data. \n
Ingestion points: Untrusted data enters the agent context via the read_file, read_lines, or http_get preset tools documented in references/step-types.md. \n
Boundary markers: Documentation examples do not include delimiters or instructions for the agent to ignore embedded commands within the ingested data. \n
Capability inventory: The agent tools include high-impact capabilities such as bash, save_content, exec_python, and exec_ts. \n
Sanitization: The provided examples do not demonstrate sanitization or validation of ingested content before it is processed by the LLM.

osmedeus-expert