description: Creates concise, structured scientific documents with LaTeX equations, integrated figures, and clear technical writing. Use when documenting results, writing research summaries, or creating technical reports. name: scientific-writing priority: high

Scientific Writing

Creates publication-quality markdown documents with mathematical rigor and visual clarity.

Core Principles

1. •Concision: Every sentence must earn its place
2. •Structure: Logical flow with clear hierarchies
3. •Precision: Technical accuracy over verbosity
4. •Integration: Seamlessly embed figures and equations
5. •Clarity: Complex ideas expressed simply

Document Structure

Standard Template

# [Descriptive Title]

## Summary
[2-3 sentences: what was done, key finding, significance]

## Methods
[Essential details only - enough to reproduce]

## Results
[Observations → Data → Interpretation]

### [Subsection 1]
[Finding with supporting figure/equation]

### [Subsection 2]
[Finding with supporting figure/equation]

## Discussion
[Implications, limitations, future directions - 1 paragraph max]

Section Guidelines

Summary (3-5 lines)

• •First sentence: what you investigated
• •Second sentence: how you investigated it
• •Third sentence: what you found
• •NO background or motivation

Methods (minimal)

• •Only non-standard procedures
• •Reference standard protocols by name
• •Use lists for clarity

Results (core content)

• •Lead with observation, follow with data
• •One finding per subsection
• •Figure/equation immediately after relevant text
• •NO speculation (save for Discussion)

Discussion (1 paragraph)

• •Interpret in broader context
• •State limitations explicitly
• •Suggest next steps

Writing Style Rules

Concision Tactics

Before (wordy):

The results that we obtained from the experiment clearly demonstrate that there is a significant increase in the fluorescence intensity.

After (concise):

Fluorescence intensity increased 3.2-fold (p < 0.001).

Common eliminations:

• •❌ "It is important to note that..."
• •❌ "The results clearly demonstrate..."
• •❌ "As can be seen from the figure..."
• •❌ "In order to..." → ✅ "To..."
• •❌ "Due to the fact that..." → ✅ "Because..."

Active Voice

• •✅ "We measured force using AFM"
• •❌ "Force was measured using AFM"
• •✅ "Cells exhibited biphasic dynamics"
• •❌ "Biphasic dynamics were exhibited by cells"

Precision

Vague: "The protein concentration was high" Precise: "Protein concentration was 2.5 mg/mL"

Vague: "Cells responded quickly" Precise: "Cells responded within 30 s"

LaTeX Equations

Inline Math

Use $...$ for inline equations:

The diffusion coefficient $D = 0.43 \pm 0.05 \, \mu\text{m}^2/\text{s}$ indicates...

Display Math

Use $$ with \begin{} and \end{} for block equations:

The force balance is described by:

$$
\begin{equation}
F_{\text{total}} = F_{\text{spring}} + F_{\text{drag}} = kx + \gamma \frac{dx}{dt}
\end{equation}
$$

where $k$ is the spring constant and $\gamma$ is the drag coefficient.

Equation Environments

Common environments:

• •equation - Single numbered equation
• •align - Multiple aligned equations
• •split - Split long equations across lines

Examples:

$$
\begin{align}
\sigma_{xx} &= E \epsilon_{xx} \\
\sigma_{yy} &= E \epsilon_{yy}
\end{align}
$$

Equation Style

Good practices:

• •Define all variables immediately after equation
• •Use \text{} for words in equations: $F_{\text{max}}$
• •Use proper spacing: \, for thin space (units)
• •Use equation environment for automatic numbering

Example:

The relationship between stress $\sigma$ and strain $\epsilon$ follows:

$$
\begin{equation}
\sigma = E \epsilon
\end{equation}
$$

where $E = 5 \pm 0.3$ kPa is the Young's modulus.

Figure Integration

Workflow

1. •Generate figure during analysis (follow python-coding-standards)
2. •Save as SVG in figures/ folder
3. •Reference immediately after relevant text
4. •Caption format: statement of finding, not description

Figure Template

Force increased linearly with deformation up to 20% strain, then plateaued (Figure 1).

![Force-deformation relationship](figures/force_vs_strain.svg)
**Figure 1.** Linear stress-strain response transitions to plateau regime at ε > 0.2.

Caption Guidelines

Bad (descriptive):

Figure 1. A plot showing force on the y-axis and deformation on the x-axis.

Good (interpretive):

Figure 1. Force increases linearly with deformation until strain exceeds 0.2.

Best (finding):

Figure 1. Biphasic mechanical response reveals strain-stiffening at ε > 0.2.

Multiple Panels

For multi-panel figures:

![Analysis overview](figures/analysis_overview.svg)
**Figure 2.** Cell morphology changes with substrate stiffness. **(A)** Representative images on soft (1 kPa), medium (10 kPa), and stiff (100 kPa) substrates. **(B)** Quantification shows area increases 2.5-fold from soft to stiff. **(C)** Aspect ratio decreases with stiffness (p < 0.001, n = 150 cells per condition).

Data Presentation

Tables (use sparingly)

Only when comparing multiple conditions across multiple metrics:

| Condition | Spring Constant (pN/nm) | Relaxation Time (s) | n |
|-----------|------------------------|---------------------|---|
| Control   | 12.3 ± 1.2            | 45 ± 8             | 23 |
| Drug A    | 8.7 ± 0.9*            | 67 ± 12*           | 21 |
| Drug B    | 15.1 ± 1.8            | 38 ± 6             | 19 |

*p < 0.05 vs. control

Table caption: Place above table, descriptive

Inline Data

For single comparisons, embed in text:

Treatment increased velocity from 0.23 ± 0.03 to 0.41 ± 0.05 μm/s (p < 0.001, n = 45).

Common Patterns

Introducing Analysis

Pattern: Goal → Approach → Key result

To determine whether force depends on loading rate, we performed constant-velocity indentation at speeds from 0.1 to 10 μm/s. Peak force scaled linearly with velocity (Figure 3A), consistent with viscoelastic behavior.

Quantitative Results

Pattern: Observation → Measurement → Statistics

Cells on stiff substrates adopted elongated morphologies. Aspect ratio increased from 1.8 ± 0.2 (soft) to 3.4 ± 0.4 (stiff), a 1.9-fold change (p < 0.001, n = 120 cells, Figure 2B).

Model Validation

Pattern: Model → Prediction → Test → Result

If force transmission is primarily via focal adhesions, disrupting integrin binding should reduce traction forces. We treated cells with RGD peptide and measured a 73% reduction in traction stress (Figure 4), confirming integrin-mediated adhesion.

Checklist

Before finalizing document:

• • Summary states finding in 3 sentences
• • No sentence exceeds 25 words
• • All equations defined immediately after appearance
• • Figures referenced in text before appearing
• • Figure captions state findings, not descriptions
• • NO phrases: "clearly", "obviously", "it is important to note"
• • NO repetition between sections
• • Methods include only non-standard details
• • Discussion is ≤ 1 paragraph
• • All data includes uncertainty and sample size
• • Active voice used throughout

Quick Reference

Numbers and Units

• •Use ± for uncertainty: 5.2 ± 0.3 kPa
• •Thin space before units: 10 \, \mu\text{m}
• •Scientific notation for extremes: $3.2 \times 10^{-6}$
• •Significant figures match precision

Common Abbreviations

First use: full term (abbreviation)

Atomic force microscopy (AFM) measurements revealed...

Subsequent: abbreviation only

AFM images showed...

References

Inline citation:

Previous studies^[1,2]^ demonstrated...

End of document:

## References

1. Smith et al. (2023) Nature 612: 123-130
2. Jones et al. (2022) Science 378: 456-460

Anti-Patterns

Avoid Redundancy

Bad:

Figure 1 shows the force-distance curve. As can be seen in Figure 1, force increases with distance. The figure clearly demonstrates that...

Good:

Force increased linearly with distance (Figure 1).

Avoid Throat-Clearing

Bad:

It is important to note that our results suggest that it may be possible that cells potentially exhibit...

Good:

Cells exhibit viscoelastic behavior.

Avoid Over-Qualification

Bad:

The data appear to suggest that there might be a possible trend toward increased stiffness.

Good:

Stiffness increased 1.8-fold (p = 0.03).

Example Document

See EXAMPLE.md in this skill folder for a complete scientific results document demonstrating all principles.

Rules

image-analysis

Priority: high

Biological Image Analysis

Priority: Use MCP Tools First

Always prioritize Cellpose MCP over custom code:

• •MCPs are GPU-accelerated and tested
• •Only write scripts when explicitly requested
• •Use Python packages for tasks MCPs don't cover

Cellpose MCP Tools

Segmentation:

# Server: "user-cellpose"
CallMcpTool("user-cellpose", "segment_cells_2d",
            {"image_path": "cells.tif",
             "model_type": "cyto3",  # or "nuclei", "cpsam"
             "diameter": 30})  # 0 = auto-estimate

Models: cyto/cyto2/cyto3 (general cells), nuclei (nuclei only), cpsam (most accurate)

Key parameters: model_type, diameter, flow_threshold, cellprob_threshold, channels

Image restoration:

• •denoise_image - Remove noise
• •deblur_image - Restore blur
• •upsample_image - Super-resolution
• •restore_and_segment - Combined pipeline

Utilities: estimate_cell_diameter, load_image_info, segment_cells_batch, list_available_models

Python Packages

Use when MCPs don't cover your needs:

scikit-image

from skimage.measure import regionprops
from skimage.filters import gaussian, threshold_otsu
from skimage.morphology import opening, closing

# Measure cells after segmentation
props = regionprops(masks)
for r in props:
    area, perimeter, eccentricity = r.area, r.perimeter, r.eccentricity

Common functions: regionprops, gaussian, median, threshold_otsu, morphology operations

scipy.ndimage

from scipy.ndimage import gaussian_filter, label

smoothed = gaussian_filter(image, sigma=2)
labeled, n = label(binary_mask)

aicsimageio

from aicsimageio import AICSImage

# Load proprietary formats (ND2, CZI, LIF, LSM)
img = AICSImage("data.nd2")
data = img.get_image_data("CZYX")

Quick Reference

When to use MCPs:

• •Cell/nuclei segmentation
• •Denoise, deblur, upsample
• •Batch processing

When to write code:

• •Measurements (regionprops)
• •Statistical analysis
• •Custom filters
• •Data aggregation

project-repo

Priority: high

AI Research Workflow - Navigation Guide

This project demonstrates context engineering for AI-aided research workflows.

📖 Full documentation: See README.md for complete information about this framework.

Project Structure Quick Reference

Key locations in this repository:

• •ai-rulez/rules/ - AI guidance (see main rules for complete listing)
• •ai-rulez/agents/ - Specialized agents (e.g., literature-review)
• •ai-rulez/mcp/ - MCPs (e.g., napari-mcp, cellpose-mcp)
• •Data_analysis/ - Example datasets and data analysis workflows
• •Image_analysis/ - 23 sample microscopy images for testing
• •deep_stuff/ - Random stuff like rules of life, etc.
• •workflow.ipynb - Workflow notebook

python-coding-standards

Priority: high

Python Coding Standards

Core Principles

• •Style: Follow PEP 8
• •Execution: Keep silent - minimize print statements
• •Performance: Use vectorized operations (numpy/pandas) over loops
• •Functions: Analysis returns data only; plots accept ax argument
• •Simplicity: Write inline code for one-off logic; functions for reuse

Jupyter Notebooks

⚠️ CRITICAL: Edit .py files only, NOT .ipynb

# %% [markdown]
# # Analysis Title

# %%
# Code cells use # %%
data = pd.read_csv('data.csv')

# %%
# Maintain 5-line spacing between cells
result = analyze(data)

Notebooks sync automatically via jupytext.

Analysis vs Plotting

NEVER create plots inside analysis functions:

# ❌ BAD: Analysis creates plot
def run_simulation():
    history = compute_dynamics()
    plt.plot(history)  # DON'T DO THIS
    return history

# ✅ GOOD: Analysis returns data
def run_simulation():
    """Compute dynamics and return history."""
    return compute_dynamics()

# Plot in separate cell
history = run_simulation()
fig, ax = plt.subplots()
ax.plot(history)
plt.show()

Plot Functions

All plot functions MUST accept axes:

# ❌ BAD: Creates own figure
def plot_data(data):
    fig, ax = plt.subplots()
    ax.plot(data)
    plt.savefig('plot.png')

# ✅ GOOD: Accepts axes argument
def plot_data(data, ax):
    """Plot data on provided axes."""
    ax.plot(data)
    ax.set_xlabel('X')
    ax.set_ylabel('Y')
    return ax

# Caller controls figure
fig, ax = plt.subplots(figsize=(8, 6))
plot_data(data, ax)
plt.savefig('plot.svg')
plt.show()

Plotting Configuration

Standard setup:

import matplotlib.pyplot as plt
import minimalist

minimalist.use_style("science")
plt.rcParams.update({'figure.facecolor': 'none', 'axes.facecolor': 'none'})

# Publication figure widths
text_width = 510/72.27  # LaTeX text width in inches

Standards:

• •Use matplotlib only (no seaborn)
• •Save as SVG for vector graphics
• •Save in figures/ folder
• •Use minimalist "science" style

Code Style

Inline logic for one-off operations:

# ❌ BAD: Over-abstracted
def normalize(img):
    return (img - img.min()) / (img.max() - img.min())

for q in quadrants:
    result = normalize(q)

# ✅ GOOD: Inline for one-off
for q in quadrants:
    result = (q - q.min()) / (q.max() - q.min())

Silent execution with inline comments:

# ❌ BAD: Noisy output
print("Processing...")
for i, f in enumerate(files):
    print(f"File {i+1}/{len(files)}")
    process(f)

# ✅ GOOD: Silent with comments
for f in files:
    # Normalize → filter → threshold
    process(f)

Docstrings

Simple format with Parameters and Returns:

def analyze_data(df, threshold=0.5):
    """
    Analyze dataframe and compute metrics.

    Parameters:
    - df: pandas DataFrame with columns ['x', 'y', 'z']
    - threshold: cutoff value for filtering (default 0.5)

    Returns:
    - dict with keys 'mean', 'std', 'count'
    """
    filtered = df[df['x'] > threshold]
    return {
        'mean': filtered['y'].mean(),
        'std': filtered['y'].std(),
        'count': len(filtered)
    }

Data Analysis

Pandas best practices:

# Use vectorized operations
df['result'] = df['x'] * df['y']  # Not: df.apply(lambda row: row['x'] * row['y'])

# Method chaining for readability
result = (df
    .query('value > 0')
    .groupby('category')
    .agg({'value': ['mean', 'std']})
    .reset_index()
)

# Save as CSV
result.to_csv('output.csv', index=False)

Quick Checklist

Before committing:

• • Analysis functions return data (no plots inside)
• • Plot functions accept ax argument
• • Jupyter: edited .py not .ipynb
• • Silent execution (minimal prints)
• • Docstrings with Parameters/Returns
• • Inline logic for one-off operations
• • File order: imports → config → functions → execution

rules

Priority: critical

Main rule for the project

The agent is an expert in:

• •Data Analysis: Exploratory data analysis, statistical analysis, data visualization
• •Image Analysis: Microscopy image processing, segmentation, quantification
• •Quantitative Analysis: Numerical computation, PDEs, ODEs, and mathematical modeling
• •AI Agents & MCPs: Model Context Protocol integration, agent workflows, context engineering

Core: Non-negotiable Rules

• •Provide concise, clear explanations
• •Ask questions; don't make assumptions
• •Prioritize readable, editable, reproducible, simple, modular code
• •Write concise, technical responses

Data Analysis

• •Use pandas, numpy, scipy, matplotlib for analysis and visualization

Image Analysis Expertise

Tools and Integration:

• •napari-mcp: Interactive image visualization and analysis
• •cellpose-mcp: Cell segmentation and quantification
• •Prioritize MCP tools over custom implementations
• •Extract quantitative metrics and validate results visually
• •Document analysis steps

Mathematical Tools:

• •Use sympy-mcp for calculus, algebra, differential equations
• •Use fmcp for mathematical plotting (matplotlib, numpy)
• •Generate LaTeX equations for documentation
• •Solve systems symbolically before numerical implementation
• •Vectorized operations for performance

Workflow:

• •Save results as CSV, plots as SVG
• •Only write custom scripts when explicitly requested
• •Validate results with custom code when needed

Location of important files

• •

  Rules

  • •rules.md: Main rule file with project-wide standards and expert domains
  • •project-repo.md: Repository structure documentation and navigation guide for AI agents
  • •python-coding-standards.md: Python coding patterns, best practices, and visualization standards
  • •image-analysis.md: Image analysis workflow patterns and MCP tool prioritization
• •

  Commands

  • •git.md: Git workflow: quick commit & push, manual workflows, and status checking
  • •airulez.md: AI-Rulez workflow: generate and commit with default message.
• •

  Agents

  • •literature-review.md: Expert literature research for biophysics and cell mechanics, semantic searches across local papers, Zotero library, and web sources
• •

  MCPs

  • •napari-mcp.md: Interactive image visualization and analysis
  • •cellpose.md: Cell segmentation and quantification
  • •fmcp.md: Mathematical plotting (matplotlib, numpy, sympy)
  • •sympy-mcp.md: Symbolic mathematics and calculus
  • •claude-scientific-skills.md: Scientific computing capabilities
  • •data-forge.md: Data manipulation and analysis
  • •notion.md: Notion workspace integration
  • •zotero.md: Zotero library access for literature management
  • •cursor-ide-browser.md: Browser automation for testing

Context

architecture

Project Architecture

This project follows a modular architecture with clear separation of concerns.

Directory Structure

• •cmd/: Command-line interface and entry points
• •internal/: Private application code
• •pkg/: Public library code
• •tests/: Test files and fixtures

Key Principles

• •Dependency injection for better testability
• •Interface-based design for flexibility
• •Clear separation between business logic and infrastructure

git

Git - Complete GitHub Workflow

Repository: https://github.com/surajinacademia/AI-Aided-Research-Workflow

Command Parameters

User can type after /git to provide hints or descriptions:

• •/git quick or /git small update → Use quick workflow
• •/git new feature: add user auth → Use detailed workflow with that description
• •/git alone → Agent decides based on last commit date and change size

How to Choose Workflow

1. •Prefer user hint: If user says "quick" / "small" / "just push" → quick workflow. If they say "significant" / "new feature" / "big change" or give a description → detailed workflow.
2. •If no clear hint: Run git status and git log -1 --format=%ci to check last commit date. If last commit was within 2 days AND change set is small (few files, small diff) → quick workflow. Otherwise → detailed workflow.
3. •Default: When in doubt, use detailed workflow (propose message and ask) to avoid vague commits.

Quick Workflow (Small / Recent Updates)

When to use:

• •Changes are limited and last commit was within 2 days
• •User says "quick", "small update", "just sync", etc.
• •Typo fixes, config tweaks, small edits, daily sync

Steps:

1. •Run git status (and optionally git diff --stat to confirm scope)
2. •Stage all changes: git add .
3. •Commit with short message:
   • •If user provided text after /git, use that
   • •Otherwise use timestamp: Update: YYYY-MM-DD HH:MM
4. •Push: git push

No confirmation step - keep it fast for small changes.

Detailed Workflow (Significant Changes / New Features)

When to use:

• •New features, large refactors, or many files changed
• •User says "significant", "new feature", "release", or provides description
• •Last commit is older than 2 days and change set is non-trivial
• •Multi-file changes that you'd want to read in git log later

Steps:

1. •Run git status and git diff (or git diff --staged after temporary git add .) to see what changed
2. •Propose a commit message:
   • •If user provided description after /git (e.g., /git add auth and dashboard), use that as basis
   • •Otherwise, draft a short descriptive message following conventional commits style:
     • •feat: for new features
     • •fix: for bug fixes
     • •docs: for documentation
     • •chore: for maintenance
   • •Example: "feat: add user authentication and dashboard"
3. •Ask the user: "I'll use this commit message: <proposed message>. Confirm or edit?"
4. •Only after user confirms (or provides an edit):
   • •Stage: git add .
   • •Commit: git commit -m "confirmed message"
   • •Push: git push

Standard Git Practices

Commit Message Format

• •Use short subject line (≤72 chars)
• •Optional body for explaining "why" (separate with blank line)
• •Conventional commits: feat:, fix:, docs:, chore: for clarity

Pull Before Push

• •Before pushing, check if branch is behind remote: git status
• •If branch is ahead AND behind remote, run: git pull --rebase
• •Then push to avoid unnecessary merge commits

What Not to Commit

• •Check git status --short before staging
• •Don't stage sensitive files (secrets, .env, credentials)
• •Don't stage large data files or build artifacts
• •Rely on .gitignore for automatic exclusion

Utility Commands

Check Status Only

git status

Shows what files changed without committing.

Check Last Commit Date

git log -1 --format=%ci

Shows when the last commit was made (helps decide workflow).

Check Ignored Files

git status --ignored

Shows files that won't be committed (in .gitignore).

View Recent Commits

git log --oneline -5

Shows last 5 commits with short messages.

Stash Changes

git stash        # Save uncommitted work
git stash pop    # Restore stashed work

Use when switching branches with uncommitted changes.