Physics Solver Skills for Claude

Three exam-tuned Claude Skills that help teachers solve Indian entrance-exam physics problems accurately and reply to student WhatsApp batches with worked solutions.

Skill	For	What it does
jee-main-physics-solver	JEE Main (12th batch)	Fast, NCERT-aligned solutions. Option-letter answers for Section A, numeric for NAT.
jee-advanced-physics-solver	JEE Advanced / IIT-JEE	Symbolic derivations, multi-correct option-by-option, matrix-match, integer type.
neet-physics-solver	NEET UG / AIPMT	Short MCQ solutions ending in option letter. Heavy-weighted modern physics.

Why these skills exist

Teachers get student physics problems as WhatsApp photos all day. Plain LLMs get the answer right most of the time — but on multi-step problems they still make silent arithmetic slips, miss a factor of g, flip a sign, or pick the wrong moment-of-inertia. The cost of sending a wrong answer to a batch is reputational; the cost of re-checking every reply is the teacher's time.

These skills close the gap with a specific workflow:

1. Transcribe the problem from photo / WhatsApp forward.
2. Search the internet for existing worked solutions (Byju's, Vedantu, Doubtnut, NEETprep, official answer keys).
3. Derive independently — set up physics symbolically, solve.
4. Verify with Python + SymPy — the arithmetic step where LLMs most often fail.
5. Cross-check the two answers; iterate up to 3 times if they disagree.
6. Flag unresolved disagreements rather than silently overwriting a verified answer — because coaching keys are occasionally wrong.
7. Stage a WhatsApp-ready draft ("Concept → Formula → Steps → Answer") for the teacher to approve before sending.

Installation

Option A — Install a single skill

Zip the skill folder with a .skill extension and upload it to your Claude workspace:

zip -r jee-main-physics-solver.skill jee-main-physics-solver/

Or use the packaging script bundled with the official anthropics/skills skill-creator:

python -m scripts.package_skill jee-main-physics-solver/ ./dist/

Then open the .skill file in Claude Code / Claude.ai / Cowork — it installs with one click.

Option B — Install all three as a personal marketplace

If you run Claude Code with plugin support, point it at this repo as a local marketplace. See plugin-marketplaces docs for the current format.

How each skill behaves in a live session

When you share a physics problem (image, text, or WhatsApp forward):

1. The matching skill triggers automatically based on the exam you mention or the problem style.
2. It loads the WebSearch tool if not already available, and runs 2–5 queries against coaching sites.
3. It runs Python (SymPy) to verify arithmetic symbolically.
4. It cross-checks its answer against the internet consensus.
5. It drafts a WhatsApp message formatted for phone screens — *bold* for headings, plain-text math, Unicode Greek letters (θ, λ, ω).
6. It shows the draft and asks: "Draft ready for [batch]. Say 'send it', 'edit: ...', or 'just for me'."

It does not auto-send — the teacher always approves.

Example

From a test run, JEE Main:

Problem: Block (2 kg) projected up 30° incline at 10 m/s, μ = 0.2. Distance before it stops? (g = 10)

Internet search: 5 coaching sites agree on a = g(sinθ + μcosθ). SymPy verification: s = 125/11 − 25√3/11 = 7.4272 m. Cross-check: internet consensus and SymPy agree on Option (C) 7.4 m. ✓

Draft for batch:

Mechanics — JEE Main

Concept: Work–energy theorem on a rough incline. Key formula: v² = u² − 2as, where a = g(sinθ + μcosθ) Given: m = 2 kg, u = 10 m/s, θ = 30°, μ = 0.2, g = 10 m/s² Steps:

1. a = 10(0.5 + 0.2·0.866) = 6.73 m/s²
2. s = u²/(2a) = 100/13.46 ≈ 7.43 m

Answer: ≈ 7.4 m → Option (C)

Requirements at runtime

• A Python 3 runtime with sympy (bundled in Cowork; for Claude Code, pip install sympy).
• WebSearch tool available in the session (Claude Code: standard; Cowork: load via ToolSearch — the skill does this automatically).
• A WhatsApp connector if you want to send drafts directly (the skill stages drafts; sending is optional and teacher-approved).

Shared references

Each skill bundles four topic reference files (references/mechanics.md, references/electromagnetism.md, references/modern_physics.md, references/common_pitfalls.md). They encode the sign conventions, standard formulas, and recurring traps that cost marks — loaded only when the problem is in that topic, to keep context lean.

Shared verification helper

Each skill bundles scripts/verify.py, a small SymPy wrapper with functions for the most common problem patterns: 1D kinematics, projectile motion, inclined plane, simple pendulum, thin lens, RC transients, Bohr energies, de Broglie wavelength. Adapt in-line for novel problems.

Contributing

Issues and PRs welcome — particularly:

• Bug reports for a problem where the skill gave a wrong answer.
• Extensions to scripts/verify.py for topics it doesn't cover.
• Additional reference files for weaker topics (elasticity, fluids, error analysis).

License

MIT — see LICENSE.

Acknowledgments

Built with Anthropic's skill-creator workflow. Thanks to Byju's, Vedantu, Doubtnut, NEETprep, Allen, Akash, and other coaching sites whose public worked solutions the skills cross-reference.