How to use the Metabolic Study Planner agent skill

What the skill makes the agent do

When a request is broad, such as a metabolic flux analysis paper or a publishable idea in microbial metabolism, the skill turns it into a concrete study plan. The plan follows the shape organism plus model plus condition plus perturbation plus metric plus figure set plus claim.

The skill makes the agent choose an organism, a metabolic model (for example from the BiGG database), a condition, a perturbation, the metrics and figures, and a testable claim. It also scores the idea for feasibility and writes a short study card before generating any simulation code. It is a planning aid for constraint-based metabolic modelling, not a tool that runs experiments by itself.

• Picks one of four study types: knockout strategy for product overproduction, nutrient-condition phase map, essentiality and drug-target prioritisation, or method benchmark.
• Scores candidate ideas from 1 to 5 on model availability, runtime, interpretability, output richness, and reproducibility, and proceeds only if the total is at least 18 out of 25.
• Writes a study_card.md with research question, hypothesis, model, conditions, analyses, metrics, figures, and risks.

Install the skill from the ZIP or the Claude apps

Download the ZIP for the skill from our page. The skill folder contains a SKILL.md file. There are two ways to install it.

For Claude Code, unzip the download into your personal skills folder at ~/.claude/skills/ so the skill is available in every project. To limit it to one project, unzip it into that project's .claude/skills/ folder instead.

In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings, then Capabilities, then Skills, then upload

How to trigger the skill with natural requests

You do not need a special command. The skill is matched when your request is about metabolic studies, metabolic engineering, FBA, COBRApy, BiGG models, organism selection, or generating a hypothesis. Saying that you have no idea yet is also a good trigger.

The more you state about the organism, product, or goal, the less the agent has to guess. If you give only a broad topic, the skill will propose a default organism and model.

• I want to do a metabolic flux analysis paper but have no idea which organism to use.
• Help me find a publishable idea in microbial metabolism using FBA.
• Plan a metabolic engineering study to increase succinate production in E. coli.
• I have no idea yet. Suggest a COBRApy study with a testable hypothesis.
• Build a study card for an organism selection in yeast fermentation.

Choose a study type and default organism

The skill matches your topic to one of four archetypes. A product overproduction topic gets a knockout strategy plan, which screens single reaction or gene knockouts and ranks them by product secretion while keeping growth. A topic about nutrient limitation, aerobic or anaerobic growth, or stress gets a two-axis nutrient phase map. A topic about antimicrobial targets, cancer metabolism, or essential genes gets an essentiality analysis. A methodological topic gets a benchmark that compares FBA, pFBA, loopless FBA, and FVA.

If you do not name an organism, the skill starts from a low-risk default. E. coli K-12 with iJO1366 or a core model is the recommended choice for fully autonomous first runs because it is fast and interpretable.

• E. coli K-12: iJO1366 or core model (fast, standard for FBA papers)
• S. cerevisiae: iMM904 (fermentation and product-yield studies)
• Human metabolism: Recon3D (disease metabolism, but larger and harder)
• M. tuberculosis: iNJ661 (essentiality and drug-target hypotheses)

What you need before running the plan

The skill plans code that uses COBRApy, so you need Python with COBRApy installed for the downstream work. The SKILL.md file refers to COBRApy scripts, but the file itself installs nothing, so you set up Python yourself.

You also need access to model files, such as BiGG, SBML, or JSON models. If the full model is unavailable, the generated script can run on a minimal fallback toy model instead.

• Python with COBRApy for the code the plan describes
• Model files in BiGG, SBML, or JSON format, or the fallback toy model
• A basic understanding of FBA and COBRApy, since the plan uses dense domain terms

Safety notes: scripts, network, and what the outputs mean

This skill does not run scripts and does not use the network. Its outputs are in-silico predictions from genome-scale models. They are not validated biology, and any drug-target claim needs experimental follow-up.

The skill tells the agent not to overclaim. Results are phrased as model-based predictions, and paper conclusions must state that they are constraint-based computational predictions requiring wet-lab validation. The skill is not meant for wet-lab or clinical decisions.

• Runs scripts: no
• Uses network: no
• Outputs are model-based predictions, not experimental results
• Not for clinical or wet-lab decisions

Tips for getting a useful plan

Give the skill a specific product, organism, or question if you have one. Use the study card as a checkpoint before any code is generated, and check that its hypothesis is falsifiable. If your idea scores below 18 out of 25, narrow the organism, product, or perturbation space rather than adding scope.

If you already have a defined model and question and only need code, use a sibling skill instead: gsmm-builder, fba-simulator, or flux-analyzer. The skill is also used inside AutoResearchClaw, where it ties hypotheses to a named model and analysis.

• Name the product (for example succinate or ethanol) if you are doing metabolic engineering.
• Ask for the study card first and review the risks section.
• Narrow the study if the feasibility score is under 18.
• Use gsmm-builder, fba-simulator, or flux-analyzer when the model and question are already set.