| name | solve-physics-problem |
| description | Solve a physics problem (text + optional figure) using the SciReasoner caption→reason→critic pipeline. Trigger when the user asks to solve, verify, or critique a physics problem that has a diagram, plot, circuit, or other figure. |
Solve Physics Problem with SciReasoner
You have access to the SciReasoner pipeline via three MCP tools:
scireasoner_solve — end-to-end (caption → reason → critic). Use this for normal solve requests.
scireasoner_caption — caption stage only (figure → structured text). Use when the user wants the figure described, not solved.
scireasoner_reason — reason stage only. Use when the user already provides a caption and only wants the derivation.
When to use
Trigger this skill when the user:
- Provides a physics problem, especially one with a figure (image attached, file path mentioned, or screenshot pasted).
- Asks to "solve", "verify", "explain", "find the answer", or "critique" a physics problem.
- Posts an image of a physics question and asks for the answer.
How to drive the tools
Standard flow
For a typical request "solve this physics problem [image attached]":
scireasoner_solve(
problem="<the user's problem text, verbatim if given; empty string if it's all in the image>",
image_path="<absolute path to the figure>",
)
The tool returns {answer, reasoning, caption}. Present the answer first, then the reasoning, and optionally the caption if the user asked about the figure specifically.
Variants
- Text-only problem: omit
image_path. The pipeline will skip captioning automatically.
- Higher accuracy at higher cost: pass
k_samples=3 to enable self-consistency voting on the reason stage. Use sparingly — each unit increase triples the reason cost.
- Quick draft: pass
use_critic=False to skip the critic+refine pass. Halves cost, slightly less reliable.
- Caption only: when the user just wants the figure described, call
scireasoner_caption instead of scireasoner_solve.
Output convention
Format your reply to the user as:
**Answer**: <final answer>
**Reasoning**:
<full chain of reasoning>
(optional, if a figure was provided)
**Figure description**:
<the structured caption>
Always preserve LaTeX as written by the model — wrap inline math in \(...\) and display math in $$...$$ only if the user is in a Markdown-rendering environment.
Notes for Claude
- The pipeline is opinionated about format: it produces
<reasoning>...</reasoning><answer>...</answer>, but the MCP wrapper extracts those for you. Just present the fields directly.
- For multiple-choice problems, the answer will be a single letter or letter combination (e.g.,
BD).
- Numerical answers usually include LaTeX-style units (
\sqrt{35}\,\mathrm{m/s}).
- If the user objects to an answer, run
scireasoner_solve again with k_samples=3 for a self-consistency vote, OR ask them to clarify what they think went wrong.