Skip to content

Latest commit

 

History

History
35 lines (27 loc) · 1.98 KB

File metadata and controls

35 lines (27 loc) · 1.98 KB
name First-time setup
description Agent-led setup of environment and HPC scheduler

Instructions for the Agent:

  1. Confirm SKILL_DIR. Ask the user to confirm the skill directory path. The default is $HOME/.claude/skills/openfe_skill. If the user provides a different path, update the SKILL_DIR="..." line in every one of these files to match:

    • 1_validate_input.md, 2_plan_rbfe_network.md, 3_submit_jobs.md, 4_gather_results.md, a_plot_correlations.md, b_setup_prescreen_by_ID_suffix.md

    Use the confirmed path as $SKILL_DIR in all commands below.

  2. Check whether setup is already done. Run:

    source $SKILL_DIR/scripts/hpc_config.sh && echo $SITE_CONFIGURED

    If the output is true, the setup is already completed; exit. If it is false or empty, continue below.

  3. Interview the user. Ask the following questions (all are required):

    • What is the command to load openfe environment (e.g. ml CEDAR; ml Python/openfe, or conda activate openfe)
    • What is the SLURM account (e.g. -A openfe), or none if not required
    • What are the GPU partition (e.g.-p gpu), GPU type (e.g. l40s) for those steps that run on gpu
  4. Edit $SKILL_DIR/scripts/hpc_config.sh based on the answers:

    • Replace the body of load_openfe_env() with the correct commands.
    • Update every SBATCH_* array: set -A, -p, --gres to match user's specifications (e.g. -A openfe, -p gpu, --gres=gpu:l40s:1 ).
    • Set SITE_CONFIGURED=true at the bottom.
  5. Validate the configuration. Run:

    bash $SKILL_DIR/scripts/check_config.sh

    All checks must pass before proceeding. If any fail, explain what the error(s) mean to user, ask user to examine $SKILL_DIR/scripts/hpc_config.sh, then ask user for input before trying to fix the error(s).

  6. Once check_config.sh reports "All checks passed", suggest user to review $SKILL_DIR/scripts/hpc_config.sh. Then proceed to Step 1 (Input Validation).