Data Analysis

umap-tsne-analysis

Performs sample-level dimensionality reduction and visualization on abundance or OTU-style matrices with a companion group file, generating UMAP and/or t-SNE coordinates and scatter plots for group separation assessment.

85100Total Score

Core Capability

85 / 100

Functional Suitability

11 / 12

Reliability

10 / 12

Performance & Context

7 / 8

Agent Usability

13 / 16

Human Usability

7 / 8

Security

11 / 12

Maintainability

10 / 12

Agent-Specific

16 / 20

Medical Task

22 / 25 Passed

90Both UMAP and tSNE on OTU matrix with 3 groups

5/5

87UMAP-only with custom column names and no normalization

5/5

84Single-group input

4/5

85Both methods with 120-second timeout

4/5

81Partial sample ID overlap between group file and matrix

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 fabrication of statistical values; dimensionality reduction outputs computed by Rtsne and umap packages on user data
Practice Boundaries	PASS	No medical diagnoses; skill produces visualization coordinates and plots, not clinical conclusions
Methodological Ground	PASS	t-SNE with perplexity and UMAP with n_neighbors are standard approaches; Hellinger normalization appropriate for abundance data; single-group hard stop is methodologically correct
Code Usability	PASS	Syntactically correct; both methods use well-maintained R packages; timeout wrapper with with_timeout() is correct; no infinite loops; clean exit codes

Core Capability85 / 100 — 8 Categories

Functional Suitability

Covers both methods independently and together, custom column resolution, normalization control, and timeout; minor gap: no explicit handling documented for datasets where n_samples < perplexity

11 / 12

92%

Reliability

Seven SKILL_* codes including SKILL_TIMEOUT; conditional R.utils dependency is correct; validate_groups() enforces design requirements; one gap: SKILL_SAMPLE_MISMATCH does not list specific mismatched IDs

10 / 12

83%

Performance & Context

SKILL.md is 314 lines within budget; references directory for algorithm details; version-pinned dependency baseline in dependencies.lock.tsv; one minor gap: no per-dataset-size timeout guidance

7 / 8

88%

Agent Usability

Agent Response Contract specifies 5 outputs; When to Read External Files table present; three Examples cover common scenarios; one gap: error messages lack specific ID lists; feedback for single-group case does not identify the group

13 / 16

81%

Human Usability

Natural trigger language; When Not to Use clear; prerequisites section clearly warns packages must be installed; Forgiveness: strict validation correct; one gap: package-before-file error ordering documented in troubleshooting note

7 / 8

88%

Security

No hardcoded credentials; no eval/exec; norm_method passed to vegan::decostand() but limited to documented string choices; one minor gap: norm_method not validated against allowed vegan choices

11 / 12

92%

Maintainability

Modular 7-script structure; dependency lock file adds reproducibility; testthat coverage mentioned in Implementation Checklist; one gap: install_dependencies.R not validated as part of polished package; no version lock integration in main flow

10 / 12

83%

Agent-Specific

Precise trigger language for UMAP and t-SNE; SKILL.md under 500 lines; idempotent via fixed seed; escape hatches present; conditional R.utils dependency is elegant; minor gap: no documentation of recommended perplexity ranges for different sample counts

16 / 20

80%

Core Capability Total85 / 100

Medical TaskExecution Average: 85.4 / 100 — Assertions: 22/25 Passed

Canonical

Both UMAP and tSNE on OTU matrix with 3 groups

5/5 ✓

Variant A

UMAP-only with custom column names and no normalization

5/5 ✓

Edge

Single-group input

4/5 ✓

Variant B

Both methods with 120-second timeout

4/5 ✓

Stress

Partial sample ID overlap between group file and matrix

4/5 ✓

Canonical✅ Pass

Both UMAP and tSNE on OTU matrix with 3 groups

Full workflow documented; both methods independently executable; set.seed applied

Basic 36/40|Specialized 54/60|Total 90/100

✅A1Output routes to CLI with required parameters

✅A2Both method workflows documented and independently executable

✅A3set.seed applied for reproducibility

✅A4SKILL_* codes cover all failure modes including SKILL_TIMEOUT

✅A5Scope rejection defined for DEG, scRNA-seq, and clustering without group file

Pass rate: 5 / 5

Variant A✅ Pass

UMAP-only with custom column names and no normalization

Column resolution flexible; normalize=FALSE bypasses vegan; UMAP-only produces only UMAP outputs

Basic 35/40|Specialized 52/60|Total 87/100

✅A1sample_id_col and group_col resolution implemented

✅A2n_neighbors validated greater than 1

✅A3normalize FALSE skips vegan decostand

✅A4UMAP-only mode produces only UMAP outputs

✅A5Normalization method documented in Methods section

Pass rate: 5 / 5

Edge✅ Pass

Single-group input

SKILL_INVALID_PARAMETER correctly triggered; rationale documented; group name not shown in error

Basic 34/40|Specialized 50/60|Total 84/100

✅A1SKILL_INVALID_PARAMETER triggered for single-group input

✅A2Rationale for hard stop documented in SKILL.md

✅A3At least 2 samples per group requirement enforced

❌A4Error message identifies which group is the only one found

✅A5Scope rejection for analyses without group file defined

Pass rate: 4 / 5

Variant B✅ Pass

Both methods with 120-second timeout

SKILL_TIMEOUT documented; R.utils conditionally required; timeout wraps both methods; no size-based guidance

Basic 34/40|Specialized 51/60|Total 85/100

✅A1SKILL_TIMEOUT error code documented in Error Handling table

✅A2R.utils only required when timeout greater than 0

✅A3Timeout wraps both UMAP and tSNE workflows

❌A4Recommended timeout values for different dataset sizes documented

✅A5Timeout 0 disables the limit and is documented

Pass rate: 4 / 5

Stress✅ Pass

Partial sample ID overlap between group file and matrix

SKILL_SAMPLE_MISMATCH triggered; IDs not listed; zero-abundance removal documented

Basic 33/40|Specialized 48/60|Total 81/100

✅A1SKILL_SAMPLE_MISMATCH triggered for ID mismatch

✅A2No fuzzy-matching attempted

❌A3Mismatched sample IDs listed in error message

✅A4Zero-abundance sample removal post-alignment documented

✅A5Package-before-file error ordering documented

Pass rate: 4 / 5

Medical Task Total85.4 / 100

Key Strengths

Conditional R.utils dependency (only required when timeout > 0) demonstrates thoughtful dependency management
Version-pinned dependency baseline in dependencies.lock.tsv enables reproducibility audits across environments
Flexible column name resolution via resolve_sample_id_col() and resolve_group_col() reduces input format rigidity
Correct methodological hard stop for single-group inputs with documented rationale
SKILL_TIMEOUT with R.utils setTimeLimit() wrapper provides reliable execution bounds for large datasets