The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the nel CLI tool to build configurations, list tasks, check status, and run evaluations (nel skills build-config, nel run, nel status). It also provides ssh command templates for users to manually inspect logs on remote SLURM clusters.\n- [EXTERNAL_DOWNLOADS]: The skill is designed to download model-specific scripts (such as reasoning parsers) from HuggingFace using curl as part of the pre_cmd configuration for model deployment.\n- [REMOTE_CODE_EXECUTION]: The assistant generates and instructs the execution of shell scripts via the pre_cmd and post_cmd fields in the NEL configuration. It correctly advises the user to set the NEMO_EVALUATOR_TRUST_PRE_CMD environment variable to enable this functionality safely within the tool's framework.\n- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection (Category 8) because it ingests data from external model cards to populate configuration fields that influence execution logic.\n
Ingestion points: External model cards retrieved via WebSearch from HuggingFace or NVIDIA repositories in SKILL.md (Step 3).\n
Boundary markers: Absent; the instructions do not specify the use of delimiters or warnings when processing the external model card text.\n
Capability inventory: The skill possesses the ability to execute CLI commands and generate scripts that are later executed during the nel run process.\n
Sanitization: Absent; the instructions prompt the agent to extract and apply all relevant configurations directly from the found model card text without specific validation steps.

nel-assistant