@researai/deepscientist 1.5.17 → 1.6.0

package/src/skills/science/references/packages/cesm.md

@@ -0,0 +1,80 @@
+# CESM (Community Earth System Model)
+
+## Catalog
+
+- Package id: `cesm`
+- Domains: `workflow_provenance`
+- Tags: `climate-modeling`, `earth-system`, `geoscience`, `hpc`, `coupled-model`
+- Knowledge URL: https://github.com/skilled-scipkg/CESM
+- Source archive URL: https://github.com/skilled-scipkg/CESM/archive/refs/heads/cesm3.0-alphabranch.zip
+- Upstream project URL: https://github.com/ESCOMP/CESM
+- Homepage: http://www.cesm.ucar.edu/
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+CESM is a coupled Earth system modeling framework for simulating climate dynamics across atmosphere, ocean, land, and sea ice components on HPC systems using CIME workflows.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `cesm`.
+
+## Package Check
+
+For CLI/HPC-oriented environments, check the executable or loaded module before
+running any expensive job:
+
+```bash
+command -v cesm || true
+cesm --version || true
+```
+
+If the package is available only through environment modules, record the module
+state and the exact executable path in `validation/environment/cesm_doctor.json`.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `cesm` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/chemicals.md

@@ -0,0 +1,73 @@
+# Chemicals Thermophysical Data Library
+
+## Catalog
+
+- Package id: `chemicals`
+- Domains: `workflow_provenance`
+- Tags: `chemical-engineering`, `thermodynamics`, `property-database`, `phase-equilibrium`, `thermophysical-properties`, `process-simulation`
+- Knowledge URL: https://github.com/skilled-scipkg/chemicals
+- Source archive URL: https://github.com/skilled-scipkg/chemicals/archive/refs/heads/master.zip
+- Upstream project URL: https://github.com/CalebBell/chemicals
+- Homepage: https://github.com/CalebBell/chemicals
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+Python library with on-demand databases and correlations for pure-component constants, thermophysical properties, safety information, and phase-equilibrium calculations used in chemical engineering workflows.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `chemicals`.
+
+## Package Check
+
+Use a package-specific import, executable, or module check and save the result
+under `validation/environment/chemicals_doctor.json` before treating the
+runtime as usable.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `chemicals` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/chempy.md

@@ -0,0 +1,73 @@
+# ChemPy Chemical Modeling Toolkit
+
+## Catalog
+
+- Package id: `chempy`
+- Domains: `computational_science`
+- Tags: `chemistry`, `chemical-kinetics`, `equilibrium`, `physical-chemistry`, `analytical-chemistry`, `solution-chemistry`
+- Knowledge URL: https://github.com/skilled-scipkg/chempy
+- Source archive URL: https://github.com/skilled-scipkg/chempy/archive/refs/heads/master.zip
+- Upstream project URL: https://github.com/bjodah/chempy
+- Homepage: https://github.com/bjodah/chempy
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+ChemPy is a Python chemistry library focused on reaction kinetics, equilibrium systems, and physical chemistry relations, with practical property models for aqueous and acid solutions.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `chempy`.
+
+## Package Check
+
+Use a package-specific import, executable, or module check and save the result
+under `validation/environment/chempy_doctor.json` before treating the
+runtime as usable.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `chempy` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/cirq.md

@@ -0,0 +1,73 @@
+# Cirq Quantum Circuit Framework
+
+## Catalog
+
+- Package id: `cirq`
+- Domains: `computational_science`
+- Tags: `cirq`, `quantum`, `nisq`, `quantum-circuits`, `simulation`
+- Knowledge URL: https://github.com/skilled-scipkg/Cirq
+- Source archive URL: https://github.com/skilled-scipkg/Cirq/archive/refs/heads/main.zip
+- Upstream project URL: https://github.com/quantumlib/Cirq
+- Homepage: https://quantumai.google/cirq
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+Cirq is a Python framework for building, transforming, and simulating noisy intermediate-scale quantum circuits on simulators and hardware-aware quantum devices.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `cirq`.
+
+## Package Check
+
+Use a package-specific import, executable, or module check and save the result
+under `validation/environment/cirq_doctor.json` before treating the
+runtime as usable.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `cirq` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/coffea.md

@@ -0,0 +1,73 @@
+# Coffea Collider HEP Analysis Framework
+
+## Catalog
+
+- Package id: `coffea`
+- Domains: `finite_element_engineering`, `high_energy_physics`
+- Tags: `hep`, `collider`, `columnar`, `nanoaod`, `event-analysis`, `scikit-hep`, `distributed-execution`
+- Knowledge URL: https://github.com/skilled-scipkg/coffea
+- Source archive URL: https://github.com/skilled-scipkg/coffea/archive/refs/heads/master.zip
+- Upstream project URL: https://github.com/scikit-hep/coffea
+- Homepage: https://coffea-hep.readthedocs.io
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+Coffea provides columnar high-energy collider physics analysis tools with event processing, histogramming, corrections, and scalable execution from laptops to clusters in Python.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `coffea`.
+
+## Package Check
+
+Use a package-specific import, executable, or module check and save the result
+under `validation/environment/coffea_doctor.json` before treating the
+runtime as usable.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `coffea` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/cp2k.md

@@ -0,0 +1,88 @@
+# CP2K Quantum Chemistry Suite
+
+## Catalog
+
+- Package id: `cp2k`
+- Domains: `quantum_chemistry`, `molecular_dynamics`
+- Tags: `cp2k`, `quantum-chemistry`, `electronic-structure`, `dft`, `molecular-dynamics`, `solid-state`
+- Knowledge URL: https://github.com/skilled-scipkg/cp2k
+- Source archive URL: https://github.com/skilled-scipkg/cp2k/archive/refs/heads/master.zip
+- Upstream project URL: https://github.com/cp2k/cp2k
+- Homepage: https://www.cp2k.org
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+CP2K is an atomistic simulation package for quantum chemistry and solid-state physics, supporting DFT and beyond for molecular, periodic, and materials systems with efficient parallel execution.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `cp2k`.
+
+## Package Check
+
+For Python-facing environments, start with an import/version check and then a
+minimal package-specific smoke test:
+
+```bash
+python - <<'PY'
+import importlib, json, pathlib
+package_id = 'cp2k'
+result = {"package_id": package_id, "import": "failed", "version": None, "smoke": "not_run"}
+try:
+    module = importlib.import_module(package_id.replace('-', '_').split('_jl')[0])
+    result["import"] = "passed"
+    result["version"] = getattr(module, "__version__", None)
+except Exception as exc:
+    result["error"] = repr(exc)
+pathlib.Path("validation/environment").mkdir(parents=True, exist_ok=True)
+pathlib.Path(f"validation/environment/{package_id}_doctor.json").write_text(json.dumps(result, indent=2), encoding="utf-8")
+PY
+```
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `cp2k` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/custodian.md

@@ -0,0 +1,80 @@
+# Custodian JIT Simulation Job Manager
+
+## Catalog
+
+- Package id: `custodian`
+- Domains: `quantum_chemistry`, `computational_chemistry`, `materials_science`, `workflow_provenance`
+- Tags: `materials-science`, `computational-chemistry`, `job-management`, `error-recovery`, `high-throughput`, `electronic-structure`
+- Knowledge URL: https://github.com/skilled-scipkg/custodian
+- Source archive URL: https://github.com/skilled-scipkg/custodian/archive/refs/heads/master.zip
+- Upstream project URL: https://github.com/materialsproject/custodian
+- Homepage: https://github.com/materialsproject/custodian
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+Custodian is a Python framework for just-in-time management of long-running materials and chemistry calculations, providing automated error detection, correction, and restart workflows for high-throughput simulation campaigns.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `custodian`.
+
+## Package Check
+
+For CLI/HPC-oriented environments, check the executable or loaded module before
+running any expensive job:
+
+```bash
+command -v custodian || true
+custodian --version || true
+```
+
+If the package is available only through environment modules, record the module
+state and the exact executable path in `validation/environment/custodian_doctor.json`.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `custodian` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.

package/src/skills/science/references/packages/dart.md

@@ -0,0 +1,73 @@
+# DART Robotics Dynamics Engine
+
+## Catalog
+
+- Package id: `dart`
+- Domains: `robotics_physics`
+- Tags: `robotics`, `physics-engine`, `dynamics`, `simulation`, `multibody`, `collision`
+- Knowledge URL: https://github.com/skilled-scipkg/dart
+- Source archive URL: https://github.com/skilled-scipkg/dart/archive/refs/heads/main.zip
+- Upstream project URL: https://github.com/dartsim/dart
+- Homepage: https://dart.readthedocs.io
+- Catalog source: FermiLink skilled-scipkg, commit `93f089a333a43089fb1a08a73c37d05fd6683214`
+
+## When To Consider
+
+DART is a C++20 robotics and animation dynamics engine with stable articulated rigid-body simulation, collision and constraint handling, and Python bindings via dartpy.
+
+## DeepScientist Runtime Rule
+
+This card is package knowledge and routing context only. It does not mean the
+solver, Python module, CLI binary, compiled backend, license server, dataset, or
+HPC module is installed in the active environment. Before computed work, use
+`bash_exec(...)` to perform an import, executable, version, and smoke-test check
+appropriate for `dart`.
+
+## Package Check
+
+Use a package-specific import, executable, or module check and save the result
+under `validation/environment/dart_doctor.json` before treating the
+runtime as usable.
+
+Record the result with `artifact.science(...)` as `science.package_check`. Use
+`status="passed"` only when the environment can run at least a minimal smoke
+path. Use `status="failed"` or `status="blocked"` when the check explains why
+execution cannot proceed.
+
+Generated import or executable names are starting points. If `dart` uses
+a different Python module, CLI binary, environment module, container, or wrapper
+script, adjust the check before concluding the solver is unavailable.
+
+## Expected Science Nodes
+
+- `science.package_check` for import/executable/version/smoke-test evidence
+- `science.computational_run` for solver execution, simulation, fitting, or numerical computation
+- `science.dataset_analysis` when the task primarily analyzes existing data
+- `science.parameter_sweep` when varying parameters, inputs, models, or solver settings
+- `science.validation_result` for convergence, units, schema, controls, or correctness checks
+- `science.claim` only after evidence paths or related nodes support the claim
+
+## Evidence Path Conventions
+
+- `simulations/inputs/` for generated or selected solver inputs
+- `simulations/logs/` for stdout, stderr, scheduler logs, or solver logs
+- `simulations/outputs/` for structured run outputs
+- `analyses/scripts/`, `analyses/logs/`, and `analyses/outputs/` for dataset analysis
+- `validation/environment/` for package checks
+- `validation/runs/` for convergence, unit, schema, or correctness sidecars
+- `figures/` for derived visualizations
+
+## Validation Checklist
+
+- Record package version, executable path, backend, module state, or container image when relevant.
+- Preserve input files and parameters that define the scientific state.
+- Capture units, coordinate conventions, timestep/mesh/basis/model settings, seeds, and convergence criteria when applicable.
+- Validate output schema and important physical or statistical invariants before recording a computed claim.
+- Link claims to run, analysis, sweep, and validation nodes rather than relying on prose.
+
+## Common Pitfalls
+
+- Do not treat the package card or knowledge URL as runtime availability.
+- Do not weaken solver tolerances, physical models, dataset filters, or convergence criteria to make a run pass unless the change is explicitly part of the scientific question.
+- Do not call a value `computed` unless the corresponding run or analysis happened in the current quest and evidence paths are recorded.
+- Do not copy package knowledge-base material into the quest without preserving its source and license context.