Data Analysis
svm-model-importance-analysis
Train support vector machine classifiers and assess feature importance via recursive feature elimination (RFE). Inputs: feature matrix, binary class labels. Outputs: trained SVM model, RFE importance ranking, optimal feature subset, ROC performance curve.
93100Total Score
Core Capability
94 / 100
Functional Suitability
11 / 12
Reliability
10 / 12
Performance & Context
8 / 8
Agent Usability
15 / 16
Human Usability
8 / 8
Security
12 / 12
Maintainability
11 / 12
Agent-Specific
20 / 20
Medical Task
20 / 20 Passed
95Bundled binary SVM-RFE run
4/4
94Tolerance-rule custom run
4/4
91Missing plot-only bundle
4/4
92Successful plot-only regeneration
4/4
89Corrupt plot-only bundle
4/4
Veto GatesRequired pass for any deployment consideration
Skill Veto✓ All 4 gates passed
✓
Operational Stability
System remains stable across varied inputs and edge cases
PASS✓
Structural Consistency
Output structure conforms to expected skill contract format
PASS✓
Result Determinism
Equivalent inputs produce semantically equivalent outputs
PASS✓
System Security
No prompt injection, data leakage, or unsafe tool use detected
PASSResearch Veto✅ PASS — Applicable
| Dimension | Result | Detail |
|---|---|---|
| Scientific Integrity | PASS | No fabricated scientific claims or invented performance numbers were observed in any output. |
| Practice Boundaries | PASS | The skill stayed within a computational feature-ranking workflow and made no diagnostic or prescriptive claims. |
| Methodological Ground | PASS | The documented two-class linear SVM-RFE approach matched the tested binary dataset and scope restrictions. |
| Code Usability | PASS | Canonical, custom, missing-bundle, plot-only reuse, and corrupt-bundle executions all behaved as documented during the audit. |
Core Capability94 / 100 — 8 Categories
Functional Suitability
The skill covers the main binary SVM-RFE workflow, plot-only reuse, scope limits, and reproducibility expectations clearly.
11 / 12
92%
Reliability
Input validation, path confinement, and recovery guidance are strong; the main remaining gap is limited automated coverage for negative parameter cases.
10 / 12
83%
Performance & Context
Documentation is layered well and the execution path is linear with no redundant steps.
8 / 8
100%
Agent Usability
Instructions are easy to follow and outputs are well specified; the only deduction is that one selection heuristic is implicit rather than user-controlled.
15 / 16
94%
Human Usability
Trigger language is natural, runnable examples are present, and out-of-scope boundaries are clearly called out.
8 / 8
100%
Security
Path confinement, no network activity, and no dynamic code execution were all confirmed.
12 / 12
100%
Maintainability
Scripts are modular and tests exist, but the automated suite can still cover more documented failure branches.
11 / 12
92%
Agent-Specific
Trigger precision, progressive disclosure, composability, idempotency, and escape hatches are all strong.
20 / 20
100%
Core Capability Total94 / 100
Medical TaskExecution Average: 92.2 / 100 — Assertions: 20/20 Passed
95
Canonical
Bundled binary SVM-RFE run
4/4 ✓
94
Variant A
Tolerance-rule custom run
4/4 ✓
91
Edge
Missing plot-only bundle
4/4 ✓
92
Variant B
Successful plot-only regeneration
4/4 ✓
89
Stress
Corrupt plot-only bundle
4/4 ✓
95
Canonical✅ Pass
Bundled binary SVM-RFE run
Canonical run completed successfully and produced the documented bundle, tables, plots, and session info.
Basic 38/40|Specialized 57/60|Total 95/100
✅A1Command completes successfully and logs the expected major stages
✅A2The documented analysis artifacts are created in the requested output directory
✅A3The selected feature table is consistent with the ranking output
✅A4The workflow stays within the documented scope
Pass rate: 4 / 4
94
Variant A✅ Pass
Tolerance-rule custom run
Custom tolerance-rule settings executed cleanly and produced a full output set in a separate directory.
Basic 38/40|Specialized 56/60|Total 94/100
✅A1Custom feature-selection settings execute successfully
✅A2The run still produces the expected result bundle, tables, and plots
✅A3The skill remains reproducible under fixed seed settings
✅A4The workflow remains scoped to two-class SVM-RFE ranking
Pass rate: 4 / 4
91
Edge✅ Pass
Missing plot-only bundle
Missing plot-only artifact was rejected with a clear standardized error and an obvious recovery path.
Basic 37/40|Specialized 54/60|Total 91/100
✅A1Plot-only mode rejects missing result bundles before plotting
✅A2The error uses a standardized skill error code
✅A3The error is actionable and recoverable
✅A4The failure is safe and scoped
Pass rate: 4 / 4
92
Variant B✅ Pass
Successful plot-only regeneration
Existing result bundles were reused cleanly and plots were regenerated without rerunning analysis.
Basic 37/40|Specialized 55/60|Total 92/100
✅A1Existing result bundles can be reused without re-reading raw inputs
✅A2Plot regeneration succeeds with the documented minimal arguments
✅A3The command avoids unnecessary reruns of the full SVM-RFE workflow
✅A4Behavior remains deterministic on the same stored bundle
Pass rate: 4 / 4
89
Stress✅ Pass
Corrupt plot-only bundle
The corrupt RDS bundle was rejected with a standardized skill error and explicit recovery guidance.
Basic 36/40|Specialized 53/60|Total 89/100
✅A1Corrupt svm_result.rds input is rejected instead of producing undefined plots
✅A2Corrupt bundle errors use a standardized SKILL_* code
✅A3Corrupt bundle errors provide actionable recovery guidance
✅A4Failure handling remains non-destructive and within scope
Pass rate: 4 / 4
Medical Task Total92.2 / 100
Key Strengths
- The skill has a strong CLI contract with clear inputs, outputs, and scope boundaries for two-class SVM-RFE work.
- Path confinement is implemented correctly: output traversal outside the skill root was rejected with a standardized error.
- Determinism is strong: repeat canonical runs produced byte-identical ranking CSV outputs under a fixed seed.
- The implementation is modular and backed by a runnable packaged test suite that passed during the audit.