Add ziatdinovmax/SciLink to the registry

agents/ziatdinovmax__scilink/README.md

@@ -0,0 +1,85 @@
+# SciLink
+
+**LLM-powered agents for scientific research automation** — an AI research partner for
+materials science, chemistry, physics, and adjacent experimental disciplines.
+
+SciLink deploys a system of intelligent orchestrator agents that cover the full scientific
+research cycle: plan experiments, analyze multi-modal instrument data, and run
+computational simulations — all within a single, coherent session.
+
+---
+
+## Run
+
+```bash
+npx @open-gitagent/gitagent run -r https://github.com/ziatdinovmax/SciLink
+```
+
+---
+
+## What It Can Do
+
+SciLink provides three complementary orchestrator modes:
+
+### 🔬 Analyze Mode
+Multi-modal experimental data analysis powered by domain-specific skill bundles:
+- **Microscopy & Image Analysis** — particle segmentation, grain mapping, SAM-powered feature extraction
+- **Spectroscopy** — curve fitting, peak identification, XRD, Raman, EELS
+- **Hyperspectral Datacubes** — 3D datacube analysis, dimensionality reduction, component mapping
+- Generates and executes verified Python analysis code in a sandboxed environment
+- Produces reproducible session artefacts (code, plots, structured findings)
+
+### 📋 Plan Mode
+Literature-grounded experimental campaign design:
+- **Hypothesis Generation** — RAG over your papers, project notes, and prior results
+- **Bayesian Optimization** — sequential single-objective and multi-objective BO for optimal next-experiment selection
+- **Agentic Knowledge Query** — dynamically queries tabular databases and structured records without upfront schema definition
+- Knowledge-only skill bundles (no runnable code) keep planning focused on reasoning and synthesis
+
+### ⚛️ Simulate Mode
+Computational modeling with iterative structure refinement:
+- **DFT Workflows** — structure generation, VASP input preparation, INCAR validation against literature, post-run output analysis
+- **Classical MD** — LAMMPS input generation, force-field selection, trajectory analysis
+- **Structure Validation** — 3-axis visualizations, structure comparison, automated fix suggestions
+- HPC job submission and result download via SSH/SLURM backend
+
+### 🤖 Meta Orchestrator
+A top-level router that delegates to the right specialist automatically — users interact with a single interface and never switch modes manually. Context and findings flow between specialists via a delegation ledger.
+
+---
+
+## Autonomy Levels
+
+| Level | Behavior |
+|---|---|
+| **Co-Pilot** | Human reviews every step before the agent commits |
+| **Autopilot** | Agent leads; human reviews major decisions and generated outputs |
+| **Autonomous** | End-to-end execution, no human review — for headless/HPC pipelines |
+
+---
+
+## Skill Subsystem
+
+Domain experts extend SciLink without touching core agent code by contributing self-contained
+skill bundles: a `<name>.md` narrative file organized around a fixed section vocabulary
+(`Overview → Planning → Implementation → Interpretation → Validation`) plus optional
+Python helpers registered as tools. Skill-gated tool visibility keeps per-call context
+focused on the active technique.
+
+---
+
+## Install
+
+```bash
+pip install scilink            # core (analysis + planning)
+pip install scilink[ui]        # + Streamlit web UI
+pip install scilink[sim]       # + simulation dependencies (ASE, atomate2, etc.)
+pip install scilink[all]       # everything
+```
+
+Also available as an MCP server (`scilink mcp`) for integration with Claude Code and other MCP-compatible clients.
+
+---
+
+## Built with
+[gitagent](https://github.com/open-gitagent/gitagent) — a git-native, framework-agnostic open standard for AI agents.

agents/ziatdinovmax__scilink/metadata.json

@@ -0,0 +1,26 @@
+{
+  "name": "scilink",
+  "author": "ziatdinovmax",
+  "description": "LLM-powered scientific research automation: multi-modal data analysis, Bayesian experiment planning, and DFT/MD simulation via three orchestrator agents.",
+  "repository": "https://github.com/shreyas-lyzr/SciLink",
+  "version": "0.0.30",
+  "category": "research",
+  "tags": [
+    "science",
+    "materials-science",
+    "spectroscopy",
+    "microscopy",
+    "dft",
+    "bayesian-optimization",
+    "llm-agents",
+    "automation"
+  ],
+  "license": "MIT",
+  "model": "anthropic:claude-sonnet-4-6",
+  "adapters": [
+    "claude-code",
+    "system-prompt"
+  ],
+  "icon": false,
+  "banner": false
+}