Mitigates procedural p-hacking, flags silent data leakage, and provides cryptographic auditability. Equips Google Antigravity, Claude Code, Cursor, and Copilot with structured methodological guardrails, preregistered hypothesis ledgers, and tamper-evident SHA-256 manifests.
curl -fsSL https://raw.githubusercontent.com/krsdntr/OmniResearch/main/install.sh | bash
When AI coding assistants are left ungoverned, they silently modify datasets, fish for p-values, and hallucinate uncalibrated trendlines. OmniResearch establishes hard deterministic boundaries.
01_data/raw/ is monitored under continuous cryptographic integrity checks. The agent is
restricted from modifying raw files in place, with every execution validated against SHA-256 baselines.
Before executing regressions or neural sweeps, null (H₀) and alternative (H₁) hypotheses, significance α
thresholds, and power plans are committed to a local preregistration ledger in 00_meta/.
Point estimates without standard errors or 95% confidence intervals are rejected. Every reported metric must undergo ablation and sensitivity testing.
Every plot and table is cryptographically bonded to the raw inputs, transformation script, seed, and OS
specs in 05_artifacts/provenance/manifest.json.
Bridging autonomous LLM execution with the reproducible science protocols and auditability standards required by peer-reviewed research.
Replaces unmonitored exploratory runs with structured human oversight. Requires explicit researcher
digital approval before unlocking data execution pipelines and statistical inferences in
00_meta/.
Streamlines technical documentation for peer-review. Automatically compiles standardized LaTeX
(booktabs) and Markdown methodology appendices directly from cryptographic lineage manifests.
Static code analysis powered by Python Abstract Syntax Trees (AST). Automatically intercepts pre-split
transformations (e.g. StandardScaler.fit() prior to split) and temporal shuffle leaks before
execution.
Engineered for high-throughput genomic, physical, and observational datasets. Uses 8MB streaming buffers and 64MB chunked Merkle Tree roots for scalable, memory-efficient data integrity checks.
Mitigates confirmation bias and LLM sycophancy. Deploys an automated adversarial audit protocol to challenge claims via marginal p-value bounds (0.01 ≤ p < 0.05), 5% outlier sensitivity trims, and 1,000-permutation placebo tests.
How information moves through strict quarantine boundaries without leakage or retroactive tampering.
Test the tamper-detection principle live in your browser. Alter a single character in the raw patient cohort below to see how SHA-256 immediately detects tampering and invalidates the artifact receipt.
Click through each directory layer to inspect its operational role, file permissions, and scientific guardrails.
Stores formal hypothesis matrices, study objectives, and decision logs. Establishes the study protocol prior to computing statistics.
hypothesis_matrix.md: Explicit null vs. alternative hypotheses, α-thresholds, and
minimum detectable effect sizes.
decision_log.md: Chronological methodological trade-offs and modeling
decisions.
OmniResearch adapts its statistical rules depending on your study design. Click a paradigm to inspect its required checks and copy a production prompt.
Clinical dose-response, split-plot designs, laboratory biology, randomized A/B trials.
Economics, epidemiology, policy evaluations, and natural experiments.
Differential equations (ODE/PDE), molecular dynamics, stochastic kinetics.
Predictive models, tabular benchmarks, feature ablation, hyperparameter tuning.
Evidence synthesis, effect pooling, Cochrane Risk of Bias, heterogeneity.
Specialized ingestion protocols tailored to domain data structures. Coordinates, sampling frequencies, and missingness are audited before modeling.
Handles survey weights, multi-stage stratification (PSU/strata), and multivariate imputation (MICE).
Coordinate Reference System (CRS) hygiene and spatial autocorrelation (Moran's I).
RNA-Seq normalization (DESeq2), 3D PDB structure analysis, and SMILES cheminformatics.
Biomedical and physical time series (EEG/ECG). Nyquist theorem compliance and bandpass filters.
Quantitative stylometry (Burrows' Delta), topic modeling (BERTopic), and lemmatized collocations.
Topological properties: degree distributions, small-worldness, and community modularity.
Every figure and table is permanently sealed with an immutable SHA-256 fingerprint. Inspect an actual certified receipt from our clinical dose-response example.
When results are published, anyone can verify the exact computational environment and dataset that generated Figure 1:
Guarantees that raw source tables were not modified in-flight.
Proves the specific Python, R, or Julia code that ran.
Fixed random seeds (seed=42) and explicit confidence levels (95%).
{
"schema_version": "1.0.0",
"records": [
{
"artifact_id": "dose_summary.csv",
"artifact_sha256": "e5309e7567f7efdbc90aa2874efcea654d31ec4e9de031aef7e582b6a21deefd",
"timestamp_utc": "2026-09-18T09:28:17Z",
"source_script": "03_src/python/analyze_dose_response.py",
"script_sha256": "7f6c256cba08b38b6d85f3f4526645c92c0a97b99a5e0b7a883cf260d42e5851",
"input_datasets": [
{
"path": "01_data/raw/dose_response_trial.csv",
"sha256": "c88f4bd42736f26d7492c95f0eaf1b052b746c443e03300c43fc422d9bcf4936"
}
],
"parameters": {
"confidence_level": 0.95,
"model": "dose_group_summary"
},
"environment": {
"os": "Windows / Linux / macOS",
"python_version": "3.14.2"
}
}
]
}
OmniResearch integrates natively with any modern coding agent via standard discovery instruction files.
| Agentic Environment | Primary Configuration | Activation Mechanism |
|---|---|---|
| Google Antigravity | GEMINI.md + SKILL.md |
Discovered automatically; active on all data analysis and scientific inquiries. |
| Claude Code | CLAUDE.md |
Terminal CLI activation with custom slash commands (/scaffold, /toolchains).
|
| Cursor IDE | .cursorrules |
Active in Cursor Composer / Chat (Ctrl+I / Cmd+I). |
| Copilot Workspace & Codex | AGENTS.md |
Adheres to the Universal Agent Instructions standard for multi-agent workflows. |
| Open Interpreter & Aider | AGENTS.md + scripts/ |
Executes directly in local terminal environments using pure standard-library tools. |
Install globally across all workspaces or initialize directly within your current Git repository.
Installs into
~/.gemini/config/skills/research-scientist/.
Installs into
C:\Users\<User>\.gemini\config\skills\research-scientist\.
@software{omniresearch2026,
author = {krsdntr},
title = {OmniResearch: Universal Autonomous Scientific Research Workbench for AI Coding Agents},
year = {2026},
url = {https://github.com/krsdntr/OmniResearch}
}