Scholarly Specialist Orchestrator

Portable skill pack for hard problems that benefit from independent specialist panels, current evidence, and adversarial synthesis.

It normalizes a messy request, routes it to the right core and distant-field profiles, keeps the reasoning auditable, and evaluates candidate ideas against a baseline before it recommends anything.

What’s inside

• SKILL.md — main skill entrypoint and orchestration workflow.
• references/SPECIALISTS.md — human-readable specialist registry.
• references/catalog.yaml — machine-readable routing manifest.
• references/orchestration/ — input normalization, panel selection, research, synthesis, simulation, evaluation, and final report contracts.
• references/<field>/... — the specialist profiles themselves.

Install

Copy this directory into your skills root:

<skills-root>/scholarly-specialists/SKILL.md
<skills-root>/scholarly-specialists/references/...

Keep SKILL.md and references/ together.

Use

Best for:

• open-ended or cross-disciplinary problems;
• questions where first principles matter more than keyword matching;
• tasks that need current experimental research or negative results;
• candidate comparison, simulation, or baseline testing;
• disputes where assumptions need to be separated from evidence.

Example prompts:

• “Use the scholarly specialist orchestrator on this problem: …”
• “Normalize this request, then run independent core and distant specialists.”
• “Challenge the assumptions and find a non-obvious solution.”
• “Scan current research, combine compatible ideas, and test them against a baseline.”
• “Build a multidisciplinary adversarial review without majority voting.”

How it works

1. Preserve the original request and classify facts, claims, constraints, assumptions, values, and unknowns.
2. Pick a small panel of relevant core and distant specialists.
3. Run them independently on the same canonical brief.
4. Audit each brief individually.
5. Combine only compatible ideas.
6. Simulate or otherwise test the strongest candidates.
7. Report what worked, what failed, what remains uncertain, and the next useful experiment.

Specialist map

Mathematics

Logic, proof, algebra, analysis, topology, probability, numerical methods.

Physics

Classical, quantum, relativity, particle, condensed matter, and AMO physics.

Computer science

Algorithms, PL/formal methods, systems, networks, databases, security, and ML.

Chemistry

Physical and organic chemistry.

Life sciences

Molecular/cell biology, evolution/ecology, and neuroscience/cognitive science.

Engineering

Control/robotics/signal and electrical/mechanical/materials.

Social sciences

Economics, causal inference, incentives, and decision science.

Humanities

Logic, philosophy of science, ethics, and research integrity.

For the full persona list, see references/SPECIALISTS.md.

Notes

• This is a skill pack, not a project-local .omp agent tree.
• The registry is intentionally explicit so routing stays readable and reproducible.
• The main agent owns final judgment; specialists provide evidence, challenge assumptions, and propose tests.