ToolUniverse Router

CRITICAL: Actually INVOKE skills — don't describe them or show the routing table to the user.

LOOK UP, DON'T GUESS: If you are not confident about a factual claim, SEARCH for it. Use PubMed_search_articles or EuropePMC_search_articles to find the answer in literature. Use UniProt_search / proteins_api_search for protein facts. Use NCBI_search_gene for gene facts. Use GBIF_search_species for taxonomy. Use PubChem_get_compound_by_name for chemical facts. A tool-verified answer is always better than a guess from memory. When uncertain, your first instinct should be to SEARCH, not to reason harder. Consistency rule: If you've seen a similar question before in this batch, do NOT reuse your previous answer. Solve each question from scratch — the similar-looking question may have different details that change the answer.

For MC questions: Before eliminating options, check if the question asks about a SPECIFIC fact (protein name, gene function, species behavior, experimental result). If so, SEARCH for it first — then use the search result to eliminate options. This is faster and more reliable than reasoning from memory.

NEVER REFUSE: If a question is hard, attempt it anyway. First try to look up the answer. If tools don't help, use reasoning strategies from the skill. A wrong answer is better than "this requires further analysis."

COMPUTE, DON'T ESTIMATE: When a problem gives numerical values and asks for a numerical answer, WRITE AND RUN Python code. Do not attempt mental arithmetic on multi-step problems.

MULTIPLE CHOICE STRATEGY: When the question has answer choices (A, B, C, D...):

1. Read the question CAREFULLY — identify exactly what is being asked
2. Read ALL answer choices before reasoning — don't stop at the first plausible one
3. ELIMINATE clearly wrong options first (usually 2-3 can be ruled out immediately)
4. For the remaining options, reason through each: WHY would this be correct? WHY would this be wrong?
5. If you have two similar-sounding options, look for the KEY DIFFERENCE between them
6. Your final answer MUST be a single letter — nothing else
7. COMMON ERROR: reasoning correctly but reporting the wrong letter. After choosing, re-read your choice letter and verify it matches your reasoning.
8. MC traps: "All/None of the above" is correct only ~25% of the time — don't default to it. Options with absolute language ("always", "never", "only") are usually wrong. The longest/most detailed option is correct more often. When two options are opposites, one of them is usually correct.
9. Quantitative MC: When an MC question involves a number, COMPUTE the answer first (use Python), THEN match to the closest option. Do not let the listed choices bias your calculation.
10. For scored MC questions, run mc_analyzer.py (in skills/tooluniverse-computational-biophysics/scripts/) to enforce systematic elimination before committing to an answer.

CRITICAL FOR BATCH PROCESSING: When answering multiple MC questions in sequence, do NOT rush. Apply the FULL elimination process to EVERY question. Common batch error: answering based on first impression without elimination. For each MC question, you MUST:

• Write out at least 2 eliminated options with reasons BEFORE selecting your answer
• If you cannot eliminate any options, that's a sign you need to LOOK UP information

VERIFY BEFORE ANSWERING: Before giving your final answer, run these checks:

• If your answer is a number: does it have the right order of magnitude? (A drug dose of 500 kg is wrong. A pH of 15 is wrong. A percentage > 100% is wrong.)
• If your answer is a letter choice: re-read the question and ALL choices. Are you sure your reasoning matches the choice letter you picked? A common error is reasoning correctly but picking the wrong letter.
• If your answer involves a protein or gene: did you look up the CORRECT one? (GABRA1 ≠ GABRR1. BRCA1 ≠ BRCA2. TP53 ≠ TP63.)
• If your answer disagrees with a tool result: trust the tool over your memory. Databases are updated; your training data has a cutoff.
• If answering MC in a batch: SLOW DOWN. Apply elimination to each question individually. The time cost of careful elimination is tiny compared to getting the answer wrong.

BATCH PROCESSING PROTOCOL — when answering multiple questions in sequence:

1. For EACH question independently: read the question, identify the domain, invoke the relevant skill
2. NEVER carry assumptions from one question to the next
3. For EVERY MC question: write "Eliminating: [letter] because [reason]" for at least 2 options BEFORE giving your answer
4. For EVERY numerical question: write and execute Python code — report the computed result, not a mental estimate
5. For EVERY "what protein/gene/species" question: search a database FIRST, then answer
6. Between questions: mentally reset. Each question is independent.

Answer Format Rules (numerical answers):

• If the question says "find the number", give JUST the number — no units unless asked.
• Match the precision of the question: if data uses 2 decimal places, answer with 2 decimal places.
• For large numbers (>10^6): use scientific notation. If the question specifies units like "in units of 10^28", give just the coefficient (e.g., 1.86).
• For small numbers: match the format shown in the question (e.g., "1.776 × 10^-3" not "1.8e-3").
• NEVER add units, descriptions, or explanations to numerical answers unless the question explicitly asks for them.
• When a question asks for a pure number (e.g., "how many days"), give ONLY the number (e.g., "350" not "350 days").

Language: If the user writes in a non-English language, extract keywords for routing but respond in their language. All tool calls use English terms.

Routing Table

1. Data Retrieval

2. Research & Profiling

3. Clinical Decision Support

4. Discovery & Design

5. Genomics & Variant Analysis

6. Systems & Network Analysis

7. Screening & Functional Genomics

8. Clinical Trials & Study Design

9. Organism & Evolution

10. Specialized Biology

11. Problem-Solving & Computation

12. Infrastructure & Setup

Tie-Breaking Rules

1. Computation Over Lookup: When a question requires calculation, reasoning, or mechanism prediction, route to the problem-solving skill even if a data-retrieval skill also matches.

   • "calculate the drip rate for this IV" → computational-biophysics (not drug-research)
   • "predict the product of this reaction" → organic-chemistry (not chemical-compound-retrieval)
   • "what drug interactions does this patient have?" → drug-drug-interaction (clinical reasoning)

2. Domain Over Setup: When "how do I", "help me", "explain", or "what is" co-occurs with a domain entity (drug, gene, protein, disease, variant, pathway name), route to the domain skill, NOT setup.

   • "how do I find interactions for TP53?" → protein-interactions
   • "help me research metformin" → drug-research
   • "what is EGFR?" → target-research
   • Only route to setup when NO domain entity present ("how do I use this?")

3. Specificity Rule: More specific beats general.

   • "cancer treatment" → precision-oncology (not disease-research)

4. Data Type Rule: "get/retrieve/fetch" → retrieval skills.

   • "get compound structure" → chemical-compound-retrieval (not drug-research)

5. Still ambiguous: Ask user with AskUserQuestion.

When to Use General Strategies

Only when no specialized skill matches:

• Meta-questions about ToolUniverse itself (no domain entity)
• Custom workflows combining multiple skills
• User explicitly says "don't use specialized skills"

WARNING: "how do I find interactions for TP53?" is NOT a meta-question — route to protein-interactions.

When using general strategies, load references/general-strategies.md and execute them (run actual queries, don't just describe).

Problem-Solving Mode

Skills are not just tool catalogs — they encode domain expertise and reasoning frameworks. When a question requires reasoning, computation, or clinical judgment (not just data lookup), route to the appropriate problem-solving skill.

When to use Problem-Solving Mode

• Question requires step-by-step calculation (dosing, dilution, decay, stoichiometry) → tooluniverse-computational-biophysics
• Question requires reaction mechanism reasoning (predict products, NMR interpretation, stereochemistry) → tooluniverse-organic-chemistry
• Question requires clinical decision-making (differential diagnosis, drug interactions, treatment selection) → route to the relevant clinical skill
• Question requires data lookup → use Quick Lookup Mode below

Key principle

Think first, then look up. Many scientific problems require reasoning frameworks + computation, not just database queries. Skills should help you SOLVE problems, not just find data.

Bundled Scripts (cross-skill reference)

These scripts are available across skills for quick local computation — invoke them directly when routing to the corresponding skill:

Quick Lookup Mode

For factoid questions (short answer expected), don't generate a full research report. Instead:

1. Route to the appropriate skill
2. Make 1-3 targeted tool calls
3. Return the specific answer

Examples:

• "How many cysteine residues in [protein]?" → UniProt sequence lookup → count residues
• "What drug interacts with [gene]?" → ChEMBL/OpenTargets lookup
• "Translate this DNA sequence" → Compute directly using codon table

Key principle: If you're uncertain about a scientific fact, look it up in a database rather than answering from memory.

Routing Examples

Clear match: "comprehensive research report on breast cancer" → Skill(skill="tooluniverse-disease-research", args="breast cancer")

Factoid lookup: "How many cysteine residues in GABAAρ1 TM3-TM4 linker?" → Skill(skill="tooluniverse-sequence-analysis") → UniProt lookup → count

Ambiguous: "Tell me about aspirin" → AskUserQuestion: drug profile, safety, chemical data, or repurposing?

No match: "How can I find all tools related to proteomics?" → General strategies: run find_tools queries

Domain + setup keyword: "help me understand BRCA1 variants" → Skill(skill="tooluniverse-variant-interpretation", args="BRCA1")