How to use the MAGeCK CRISPR screen analysis agent skill

What the skill makes the agent do

Without the skill, an agent has no specific procedure for CRISPR screens. With it, the agent follows a defined workflow: it counts sequencing reads against a guide library (or accepts an existing count table), runs quality-control checks on read distribution and replicates, compares a treatment group against a control group, and ranks genes by enrichment or depletion with effect sizes and false discovery rates (FDR).

The skill is built to warn about weak results rather than overstate them. It does not design guides, and it does not look up public gene essentiality scores; for existing public dependency scores, the source points to DepMap instead.

• Counts FASTQ reads with MAGeCK count, or uses an existing count table
• Validates the guide library and the sample layout, including an explicit treatment and control contrast
• Runs QC: library representation, median reads per guide, zero fractions, Gini coefficients and replicate correlation
• Runs a two-group comparison with MAGeCK test and reports genes at the requested FDR in both directions

Install the skill (ZIP download or Claude apps)

Download the skill ZIP from the page, then choose one of two routes. For a personal setup, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

Installing the skill only adds the instructions and helper. MAGeCK itself is a separate install, covered below.

• Personal: unzip into ~/.claude/skills/
• Project: unzip into .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Install MAGeCK 0.5.9.5 and its requirements

The analysis needs MAGeCK 0.5.9.5 with its compiled RRA executable, plus NumPy and SciPy. The tested versions are NumPy 1.26.4 and SciPy 1.13.1, on Python 3.11. The source's tested pattern creates an environment outside the skill folder, installs the Python packages, then downloads and unpacks the MAGeCK archive. Building from source requires a C++ compiler. Network access is needed only during installation.

After installing, put the environment's bin folder on your PATH and check that both commands work: mageck --version and mageck test --help. MAGeCK is not installed automatically by this skill, so run those checks before asking for an analysis.

• Python 3.10 or newer is required for the helper (3.11 was used for testing)
• Use an environment directory outside the skill folder
• Run mageck --version and mageck test --help before any analysis

How to trigger the skill with a natural request

You do not need special syntax. Describe the screen in plain language, name the files, and say which samples are controls and which are treated. The skill is meant for new knockout, CRISPRi or CRISPRa screens, and for enrichment or depletion contrasts. The source is specific about what the agent needs to know, so including these details helps:

• Library version and whether the screen is knockout, CRISPRi or CRISPRa
• Sample names, which are controls and which are treatment, and biological replicates
• Selection direction (enrichment or depletion) and the baseline used (for example plasmid or day-zero cells)
• Whether you have FASTQ files or an existing count table

Examples of requests

Requests that fit the skill name the data and the question. For example: "Analyze my knockout screen. FASTQ files are c1, c2 for control and t1, t2 for drug-treated, two biological replicates each. Find genes that are enriched at FDR 0.05." Another: "I have a guide count table and a library TSV. Run QC first, then test treatment against control and list the depleted genes."

If you only want an existing public essentiality score for a gene, that is outside this skill; the source directs that lookup to DepMap.

What the agent checks before trusting the results

The workflow is ordered: QC comes before statistical testing. The helper requires a headerless library TSV with guide ID, DNA sequence and gene, and a count table whose header begins sgRNA and Gene, followed by unique non-numeric sample names such as c1. It rejects ambiguous sequences and count/library mismatches. A guide with zero counts stays in the library rather than being dropped to improve QC.

The 10% zero fraction and 0.8 replicate correlation flags are review prompts, not universal pass/fail thresholds. Hits should be checked for guide concordance, negative controls and replicate consistency. The helper's FDR filter applies to completed gene results.

• Library and count table format checks
• QC on library representation, reads per guide, zero fractions, Gini and replicate correlation
• Paired tests need both lists in matching biological order
• Results are saved to a result folder with a report.json recording arguments, QC, normalization and warnings

Safety notes: scripts and network

The skill runs scripts: the bundled helper is scripts/screen_analysis.py, and it uses only the Python standard library. The skill does not use the network during analysis; network access is needed only to install MAGeCK and its dependencies. No API keys or accounts are required. The helper refuses to reuse an existing results directory, so earlier output is not overwritten.

Read the commands before running them, especially the paths to your data. The example paths in the source are illustrative and need to be replaced with your own files.

• Runs scripts: yes (scripts/screen_analysis.py)
• Uses network: no, except during installation
• No credentials needed

Tips for better results

Keep biological replicates separate. Lanes of the same library prep can be combined at counting, but independent cultures should not be summed. Choose normalization deliberately: median normalization assumes most guides are stable, while control normalization needs validated negative-control guide IDs (not gene names) and at least two controls. Read the report's applied method and warnings, because MAGeCK can switch normalization automatically for sparse data.

Be careful with interpretation. An enriched guide can mean resistance, growth advantage or a sampling artifact depending on the selection. Depletion does not automatically mean essentiality. Gene hits from a screen alone do not establish gene function or therapeutic relevance.

• Use separate biological replicates, not summed cultures
• Use control normalization only with validated negative-control guide IDs
• Check the report for warnings and the applied normalization
• Treat hits as candidates that need guide concordance and replication checks