| name | temperature-simulation-manus-familytools-fullinstr-v2-skill |
| description | Family-scoped SkillLearnBench skill for temperature-simulation using latest full-instruction v2 family injection. |
| allowed-tools | ["glm_compute_public_metrics","glm_parameter_sweep","glm_run","list_dir","netcdf_summary","nml_edit_params","nml_read_params","read_file","run_python","run_shell","write_file","write_json"] |
| metadata | {"benchmark":"SkillLearnBench","generator":"Manus","baseline":"manus_familytools_fullinstr_v2","generation_constraints":"slb_no_instance_memory_familytools_v2","allowed_gold_input":"familytools_full_instruction_v2","skill_injection_scope":"family_scoped","date":"2026-05-23"} |
Skill: Temperature Simulation
When to use this skill
Use this skill for temperature-simulation tasks to run the General Lake Model (GLM), calibrate allowed parameters within published ranges, and evaluate vertical water temperature simulations using exact RMSE metrics.
Visible input and artifact inventory
Inputs:
- Forcing data in
/root/bcs/
- Observations at
/root/field_temp_oxy.csv
- Initial config at
/root/glm3.nml
- Executable at
/root/glm
Artifacts:
- Simulation output at
/root/output/output.nc
- Evaluation metrics at
/root/metrics.json
Execution procedure for the current task
- Assess: Read instructions to identify allowed calibration parameters, their ranges, and target RMSE thresholds.
- Baseline: Use
nml_read_params to check initial parameters. Run glm_run and compute baseline metrics with glm_compute_public_metrics.
- Calibrate: Adjust allowed parameters using
nml_edit_params or glm_parameter_sweep. Ensure changes stay within published ranges.
- Evaluate: After adjustments, use
glm_compute_public_metrics to calculate exact RMSE metrics based on matched observation/simulation pairs. Do not use interpolation.
- Finalize: Once RMSE thresholds are met, ensure parameters are saved in
/root/glm3.nml. Run glm_run to generate the final /root/output/output.nc.
- Report: Save final metrics to
/root/metrics.json using write_json, ensuring all required keys are present.
Family tool routing and useful placeholder snippets
- Read Params:
nml_read_params extracts current values safely.
- Edit Params:
nml_edit_params modifies parameters. Example: nml_edit_params(nml_path="/root/glm3.nml", params={"Kw": 0.3}).
- Run Model:
glm_run executes the simulation.
- Compute Metrics:
glm_compute_public_metrics calculates required RMSE values directly, handling exact datetime and rounded-depth merging.
- Sweep Params:
glm_parameter_sweep systematically explores the parameter space.
Validation checks before final submission
/root/glm3.nml contains modifications ONLY to explicitly allowed parameters.- Final parameter values are strictly within published ranges.
/root/output/output.nc exists and matches the final parameter set./root/metrics.json exists, contains all required keys, and satisfies RMSE thresholds.
Common failure modes and repair actions
- Out of Bounds: Parameter outside allowed range. Repair: Use
nml_edit_params to correct the value.
- Unauthorized Changes: Modifying unlisted parameters. Repair: Revert unauthorized changes in
/root/glm3.nml.
- Incorrect Metrics: Using custom scripts instead of exact matching. Repair: Rely exclusively on
glm_compute_public_metrics.
- Missing Output: Failing to run the model after final adjustment. Repair: Execute
glm_run before writing metrics.
Finalizer capsule
If nearing the maximum step limit without finding optimal parameters, restore the environment to a valid state. Ensure /root/glm3.nml contains parameters within allowed ranges. Run glm_run to generate /root/output/output.nc and use glm_compute_public_metrics to generate /root/metrics.json. Submitting a valid, complete set of artifacts with suboptimal performance is better than an invalid state.
