Data Analysis

univariate-multivariable-cox-regression

Use when running prognostic survival analysis on a clinical cohort with time-to-event data to estimate univariate and multivariable Cox proportional hazards models, export result tables, and generate forest plots. NOT for: nomogram construction, calibration curves, time-dependent ROC analysis, or model training/feature selection beyond the built-in univariate screening rule.

86100Total Score

Core Capability

92 / 100

Functional Suitability

12 / 12

Reliability

10 / 12

Performance & Context

7 / 8

Agent Usability

15 / 16

Human Usability

7 / 8

Security

11 / 12

Maintainability

12 / 12

Agent-Specific

18 / 20

Medical Task

23 / 25 Passed

86Basic analyze command on sample clinical data

5/5

84Forest plot generation from existing result table

5/5

81skip_univariate mode and insufficient event count

4/5

83Timeout parameter and multiple feature columns

5/5

78Missing required columns and Cox fitting failure

4/5

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

PASS

Research Veto✅ PASS — Applicable

Dimension	Result	Detail
Scientific Integrity	PASS	No fabricated statistics; Cox PH models use standard survival package; HR tables export computed values only; three canonical references cited
Practice Boundaries	PASS	No diagnostic conclusions; skill is scoped to HR table and forest plot generation; When Not to Use section explicitly excludes clinical decision tools
Methodological Ground	PASS	Univariate screening with p<0.05 threshold then multivariable model is standard practice; fallback to all features if fewer than 3 significant is appropriate and documented; no causal claims made
Code Usability	PASS	Script syntax valid; set.seed(options$seed) present; modular 8-file structure; dependency check fires with named packages; no logic errors detected in code review

Core Capability92 / 100 — 8 Categories

Functional Suitability

Three subcommands (analyze, forest-plot, multi-forest-plot) cover full workflow; all documented use cases covered with examples

12 / 12

100%

Reliability

SKILL_* codes comprehensive; set.seed() present; Cox fitting failure handled with SKILL_INVALID_PARAMETER; minor gap: no explicit warning when fallback logic triggers

10 / 12

83%

Performance & Context

SKILL.md is 327 lines — slightly over the 500-line threshold but within acceptable range; modular references and scripts well-utilized

7 / 8

88%

Agent Usability

Subcommand routing is clear; output columns documented with field descriptions; minor gap: forest-plot required columns may confuse if analyze was run with skip_univariate

15 / 16

94%

Human Usability

When to Use / When Not to Use sections well-structured with typical user request phrasings; high discoverability

7 / 8

88%

Security

No credentials or secrets; set.seed prevents side-channel variance; file path inputs accepted without explicit sanitization

11 / 12

92%

Maintainability

Eight-file modular structure; smoke test script (run_smoke_test.R) and shell wrapper included; test data bundled; references documented

12 / 12

100%

Agent-Specific

Trigger description precise with NOT-for exclusions; progressive disclosure excellent; escape hatches clearly stated; minor composability friction from xlsx output format dependency

18 / 20

90%

Core Capability Total92 / 100

Medical TaskExecution Average: 82.4 / 100 — Assertions: 23/25 Passed

Canonical

Basic analyze command on sample clinical data

5/5 ✓

Variant A

Forest plot generation from existing result table

5/5 ✓

Edge

skip_univariate mode and insufficient event count

4/5 ✓

Variant B

Timeout parameter and multiple feature columns

5/5 ✓

Stress

Missing required columns and Cox fitting failure

4/5 ✓

Canonical✅ Pass

Basic analyze command on sample clinical data

openxlsx and survival packages missing; --help renders correctly; SKILL_PACKAGE_NOT_FOUND fires correctly naming packages; set.seed confirmed

Basic 34/40|Specialized 52/60|Total 86/100

✅A1--help renders three subcommand descriptions without error

✅A2SKILL_PACKAGE_NOT_FOUND fires with named package list

✅A3set.seed() present in main.R

✅A4Modular 8-file script structure confirmed

✅A5No security issues in script code

Pass rate: 5 / 5

Variant A✅ Pass

Forest plot generation from existing result table

Subcommand routing confirmed; forestplot package dependency check fires; required plot table columns documented

Basic 33/40|Specialized 51/60|Total 84/100

✅A1forest-plot and multi-forest-plot subcommands routed correctly in main.R

✅A2Required plot table columns documented

✅A3SKILL_PACKAGE_NOT_FOUND fires for forestplot dependency

✅A4--plot_save argument documented as required for plot commands

✅A5Plot dimension customization arguments documented

Pass rate: 5 / 5

Edge✅ Pass

skip_univariate mode and insufficient event count

skip_univariate argument documented; event-count threshold documented; fallback logic documented; runtime verification blocked by missing packages

Basic 32/40|Specialized 49/60|Total 81/100

✅A1--skip_univariate argument documented for bypassing univariate screening

✅A2Minimum 2 events and 10 complete samples requirement documented

✅A3Multivariable fallback logic documented

❌A4Runtime behavior of skip_univariate verified in live execution

✅A5SKILL_INVALID_PARAMETER documented for Cox model fitting failures

Pass rate: 4 / 5

Variant B✅ Pass

Timeout parameter and multiple feature columns

Timeout and seed arguments confirmed; overwrite flag documented; canonical references present; output file list comprehensive

Basic 33/40|Specialized 50/60|Total 83/100

✅A1--timeout_seconds and --seed arguments documented in both analyze and plot commands

✅A2--overwrite flag documented for non-empty output directories

✅A3Three canonical survival analysis references included

✅A4Output files for analyze command are comprehensive

✅A5Implementation Checklist confirms major design decisions

Pass rate: 5 / 5

Stress✅ Pass

Missing required columns and Cox fitting failure

SKILL_MISSING_COLUMNS and SKILL_INVALID_PARAMETER documented; time column validation requirements confirmed; runtime blocked by missing packages

Basic 31/40|Specialized 47/60|Total 78/100

✅A1SKILL_MISSING_COLUMNS documented for absent required columns

✅A2SKILL_INVALID_PARAMETER documented for Cox model fitting failure

✅A3time_col must contain finite values greater than 0 is documented

❌A4Live Cox fitting failure error handling verified

✅A5Character predictors converted to factors documented in Workflow

Pass rate: 4 / 5

Medical Task Total82.4 / 100

Key Strengths

Three-subcommand architecture (analyze, forest-plot, multi-forest-plot) cleanly separates analysis from visualization
Automated smoke test script (run_smoke_test.R) and shell wrapper enable easy regression testing
set.seed() present with configurable --seed argument ensuring full reproducibility
Multivariable fallback logic (use all features if fewer than 3 significant) is explicitly documented
When Not to Use section precisely excludes nomograms, calibration, and time-dependent ROC, preventing misuse