How to use the Phylogenetic Tree Builder agent skill

What the phylogenetics skill makes the agent do

The skill covers estimating evolutionary relationships between homologous DNA or protein sequences. It aligns sequences with MAFFT, infers a tree with IQ-TREE using maximum likelihood with ModelFinder and branch support, or uses FastTree as an approximate, faster alternative. It can use ETE3 for tree summaries and visualization.

The workflow also includes warnings about common mistakes. A gene tree does not by itself establish a species tree, transmission direction, or horizontal transfer, and sampling dates alone do not justify a molecular clock.

• Applies to homologous nucleotide or protein sequences, microbial gene trees and protein families.
• Reports branch support values so you can judge how reliable each branch is.
• All analyses run locally; the skill does not call a remote inference API.

Install the skill from the ZIP

Download the ZIP for this skill from this page and unzip it. To make it available across your projects, place the folder in ~/.claude/skills/. To limit it to one project, place it in that project's .claude/skills/ folder.

Alternatively, in the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.

• Personal install: ~/.claude/skills/phylogenetics/
• Project install: .claude/skills/phylogenetics/ inside your project

Install the external tools and Python packages

The skill orchestrates programs you must install yourself. It cannot run without MAFFT and an IQ-TREE or FastTree executable. The source's illustrative setup uses conda for the native tools and uv for a Python 3.12 environment with ETE3.

Verify the executable versions after installation. The helper defaults to the names iqtree3 and FastTree; if you have IQ-TREE 2, pass --iqtree-bin iqtree2. Version 2 compatibility is source-checked but was not runtime-tested.

• Required: MAFFT, and IQ-TREE 3 (or IQ-TREE 2 via --iqtree-bin) or FastTree.
• Optional: Python 3.12 with ete3==3.1.3, numpy and six for tree summaries; PyQt5 for rendering.
• Optional: trimAl for trimming alignments.

Trigger the skill with natural requests

You do not need special syntax. Describe the biology and what you want back. Requests that name sequences, a protein family, or a comparison across organisms match the skill's purpose best. Mention whether the sequences are DNA (nt) or protein (aa), since the helper needs that to select the correct settings.

Be clear about inputs you already have, such as an existing alignment, so the agent does not realign it unnecessarily.

• "Build a maximum likelihood gene tree from these 12 nucleotide sequences of the rpoB gene and report branch support."
• "Align these protein sequences with MAFFT and make a FastTree tree for a quick look."
• "I already have a reviewed alignment in aligned.fasta. Infer a tree from it without realigning."

Run an analysis and read the output

The helper script is scripts/phylogenetic_analysis.py. The skill uses IQ-TREE's standard command with ModelFinder (-m MFP), 1000 ultrafast bootstrap replicates (-B 1000) and SH-aLRT with 1000 replicates (--alrt 1000). Results are written under the output directory you choose.

Read the .iqtree and .log files, not only the picture. In the default combined test, IQ-TREE writes slash-delimited SH-aLRT/UFBoot labels, so check their order in the report. UFBoot of 95 or more and SH-aLRT of 80 or more are common screening thresholds, not proof of correctness.

• Primary outputs: <prefix>.treefile (IQ-TREE) or <prefix>.tree (FastTree), plus the alignment.
• IQ-TREE files: .iqtree, .log and .ckp.gz checkpoints.
• .model.gz caches model-selection work and is not a parameter report.
• Use a new output prefix when data or model assumptions change; --redo replaces results.

Prepare good input and keep a record

Use unique, whitespace-free FASTA IDs and keep a separate sample metadata table. The helper rejects empty or all-missing records and Newick punctuation in IDs. Keep the original sequences, accession versions, filters, alignment and exact commands so the analysis can be repeated.

Coding DNA needs a codon-aware alignment if you use a codon model, because ordinary MAFFT nucleotide alignment does not preserve reading frames automatically. Trimming is optional; a failed trim must stop the step rather than silently passing the untrimmed file on as trimmed.

• Select homologs with appropriate taxon coverage and check for paralogs and contamination.
• Inspect alignment lengths, gaps, composition and taxon identities.
• Single-thread runs make results easier to reproduce.

Safety notes: scripts and network

The skill runs scripts, so review the helper and the commands it launches before using it on sensitive data. It does not use the network: the source states that all analyses are local and that no HTTP API or remote inference endpoint is invoked. Network access is needed only for installing the external tools and packages.

• Runs scripts: yes. Uses network: no.
• The helper preserves the supplied tree orientation unless you pass --outgroup or --midpoint for display; neither creates a dated phylogeny or overwrites the original tree.
• Choose outgroups on biological grounds; midpoint rooting is only a heuristic.