Data Analysis

cerna-analysis

Use when building a ceRNA regulatory network from a key gene list by combining bundled miRNA-mRNA and miRNA-lncRNA database files, with flat-file CSV exports and PDF visualization in a single output directory. NOT for: differential expression, single-cell analysis, enrichment analysis, or workflows without a key gene list.

87100Total Score

Core Capability

87 / 100

Functional Suitability

11 / 12

Reliability

11 / 12

Performance & Context

7 / 8

Agent Usability

15 / 16

Human Usability

7 / 8

Security

12 / 12

Maintainability

11 / 12

Agent-Specific

13 / 20

Medical Task

33 / 35 Passed

91ceRNA network from gene.txt, combined mode, High strictness

5/5

88Pairwise starbase+mirdb intersection, Median strictness, fr layout, freq>=2

5/5

82All lncRNA interactions removed by high frequency threshold

4/5

88Inline comma-separated gene list, circle layout, no legend

5/5

82200-gene key list with combined mode and default kk layout

4/5

87GSEA pathway enrichment on differential expression results

5/5

89Path traversal via --key_genes ../../../etc/shadow

5/5

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 DOI, PMID, or interaction data. All miRNA-mRNA and miRNA-lncRNA interactions sourced from bundled local database files (starbase, miRDB, miRTarbase).
Practice Boundaries	PASS	No diagnostic or prescriptive conclusions. Differential expression, enrichment, single-cell, and survival workflows explicitly excluded with documented out-of-scope messages.
Methodological Ground	PASS	ceRNA network construction correctly implements miRNA sponging logic via miRNA-mRNA and miRNA-lncRNA co-targeting. Pairwise intersection modes correctly reduce to higher-confidence interactions. SKILL_INVALID_DATA raised when ceRNA layer collapses.
Code Usability	PASS	Scripts syntactically valid R. Entry-point package check for igraph before any execution. source_required_script() wraps all source() calls with SKILL_* error codes. Validation layer fully separated in validation.R.

Core Capability87 / 100 — 8 Categories

Functional Suitability

Seven dataset modes and three lncRNA strictness levels documented with corresponding database files. CLI inline gene list and file-based gene list both supported. Minor gap: no proactive warning about lncRNA frequency threshold collapsing the network for small gene lists.

11 / 12

92%

Reliability

Nine SKILL_* error codes covering file, data, dependency, parameter, timeout, and runtime failures. SKILL_INVALID_DATA correctly raised when ceRNA layer collapses. Scene Override applied: hard stop on empty lncRNA layer is correct defensive design.

11 / 12

92%

Performance & Context

SKILL.md at approximately 218 lines with well-structured progressive disclosure via references/. Methods section explains pairwise intersection edge reduction quantitatively (20-40% fewer edges). Token efficiency good.

7 / 8

88%

Agent Usability

Workflow steps clearly enumerated as four-step process. Two CLI examples with different modes. Arguments table complete. Dependency note for igraph prominently placed. Near-perfect agent usability; minor gap: no feedback format specified for post-run summary.

15 / 16

94%

Human Usability

Description uses precise trigger language ('ceRNA regulatory network', 'key gene list', 'bundled miRNA databases'). Do not use it for clearly delineated. One gap: no note on typical network size expected for a standard 10-20 gene list.

7 / 8

88%

Security

No hardcoded credentials. source_required_script() validates script existence before sourcing. File paths used for read-only operations. No eval/exec of user strings. boot_fail() handles all script loading errors cleanly.

12 / 12

100%

Maintainability

Seven scripts with excellent single-responsibility separation: validation.R, io.R, functions.R, plot_functions.R, utils.R, run_analysis.R, main.R. source_required_script() wrapper adds resilience to script loading. Test commands documented.

11 / 12

92%

Agent-Specific

Trigger description includes explicit NOT triggers. Seven dataset modes support diverse analysis strategies. Idempotency strong with fixed seed and deterministic network construction. Composability limited. Escape hatches for enrichment, DEG, single-cell, survival documented.

13 / 20

65%

Core Capability Total87 / 100

Medical TaskExecution Average: 86.7 / 100 — Assertions: 33/35 Passed

Canonical

ceRNA network from gene.txt, combined mode, High strictness

5/5 ✓

Variant A

Pairwise starbase+mirdb intersection, Median strictness, fr layout, freq>=2

5/5 ✓

Edge

All lncRNA interactions removed by high frequency threshold

4/5 ✓

Variant B

Inline comma-separated gene list, circle layout, no legend

5/5 ✓

Stress

200-gene key list with combined mode and default kk layout

4/5 ✓

Scope Boundary

GSEA pathway enrichment on differential expression results

5/5 ✓

Adversarial

Path traversal via --key_genes ../../../etc/shadow

5/5 ✓

Canonical✅ Pass

ceRNA network from gene.txt, combined mode, High strictness

All four output files generated. Combined mode loads miRNA_mRNA.csv and High strictness lncRNA file.

Basic 36/40|Specialized 55/60|Total 91/100

✅A1All four output files generated in output directory root

✅A2Fixed seed 42 applied before igraph layout computation

✅A3SKILL_DEPENDENCY_MISSING raised when igraph absent, including for --help

✅A4No fabricated miRNA-mRNA interaction data in output

✅A5Out-of-scope requests refused with documented message before execution

Pass rate: 5 / 5

Variant A✅ Pass

Pairwise starbase+mirdb intersection, Median strictness, fr layout, freq>=2

Pairwise intersection correctly reduces edge count. fr layout and freq filter applied.

Basic 35/40|Specialized 53/60|Total 88/100

✅A1Pairwise intersection loads and intersects two specified database files

✅A2lncRNA frequency threshold >= 2 applied to filter low-confidence lncRNA interactions

✅A320-40% edge reduction from pairwise vs combined mode documented

✅A4fr layout accepted as alternative network layout

✅A5starbase and mirdb database files both present in references/database/

Pass rate: 5 / 5

Edge✅ Pass

All lncRNA interactions removed by high frequency threshold

SKILL_INVALID_DATA raised correctly. Minor gap: no proactive warning about frequency threshold risk for small gene lists.

Basic 33/40|Specialized 49/60|Total 82/100

✅A1SKILL_INVALID_DATA raised when lncRNA filtering removes all interactions

✅A2Error message explains ceRNA layer has collapsed

✅A3Recovery guidance suggests lowering threshold or changing dataset/strictness

❌A4SKILL.md warns about lncRNA frequency threshold risk for small gene lists

✅A5Hard stop prevents empty-layer ceRNA network construction

Pass rate: 4 / 5

Variant B✅ Pass

Inline comma-separated gene list, circle layout, no legend

Inline gene list parsing works correctly. show_legend=FALSE hides PDF legend.

Basic 35/40|Specialized 53/60|Total 88/100

✅A1Comma-separated gene string accepted directly on CLI without a file

✅A2Blank lines and comment lines starting with # ignored in key gene file

✅A3Duplicate genes removed automatically before network construction

✅A4circle layout accepted and applied to network visualization

✅A5show_legend=FALSE hides legend in PDF output

Pass rate: 5 / 5

Stress✅ Pass

200-gene key list with combined mode and default kk layout

Large network runs with run_with_timeout(). kk layout performance warning missing from SKILL.md.

Basic 32/40|Specialized 50/60|Total 82/100

✅A1Large key gene list accepted without cap

✅A2Default timeout of 3600 seconds applies as safeguard

✅A3run_with_timeout() wraps the full analysis with elapsed timeout

❌A4SKILL.md warns that kk layout is slow for large networks with recommendation to use nicely or fr

✅A5SKILL_TIMEOUT raised when elapsed timeout exceeded

Pass rate: 4 / 5

Scope Boundary✅ Pass

GSEA pathway enrichment on differential expression results

Correctly refused. Both enrichment and DEG workflows listed as out of scope.

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

✅A1Enrichment and differential expression requests identified as out of scope

✅A2Documented out-of-scope message returned without running any analysis

✅A3No network construction attempted for enrichment request

✅A4Skill clarifies it requires a key gene list to start

✅A5Survival analysis also excluded as separate entry

Pass rate: 5 / 5

Adversarial✅ Pass

Path traversal via --key_genes ../../../etc/shadow

File not found or permission denied raises SKILL_FILE_NOT_FOUND. No code injection vector.

Basic 35/40|Specialized 54/60|Total 89/100

✅A1Non-existent or unreadable --key_genes file raises SKILL_FILE_NOT_FOUND

✅A2Path traversal attempt does not execute arbitrary code

✅A3No eval() or system() calls with user-provided file paths

✅A4Error exits cleanly with status 1 via boot_fail() or main error handler

✅A5Error message uses SKILL_* prefix format for structured parsing

Pass rate: 5 / 5

Medical Task Total86.7 / 100

Key Strengths

Excellent modular script architecture with source_required_script() providing resilient script loading and structured SKILL_* error wrapping
Seven dataset modes and three lncRNA strictness levels provide flexible interaction confidence tuning from a single local workflow
SKILL_INVALID_DATA correctly raised when ceRNA layer collapses, preventing silent construction of scientifically invalid networks
Inline comma-separated gene list CLI support enables quick testing without a file, documented alongside full file input
Methods section quantitatively documents pairwise intersection edge reduction (20-40%), setting correct user expectations