Comment Generator — How to Share It With Other Teachers
Research brief · requested 2026-04-30 · Mark Barnett
The ask
Mark wants other SFMS teachers to use the same comment-generation workflow — input a CSV from Gradebook (or Facts summary), do a voice brain dump on the class, add any documents for context, and get report-card comments in the right format. The format is the key deliverable. Teachers shouldn’t need to understand Claude or prompting.
What exists today
- Comment generator v2 — a Claude project with a system prompt. Mark pastes it into his own Claude project and runs it there. Works, but requires access to Claude + knowing how to set up a project.
- Format requirements — St. Francis specific: length, tone, no clichés, grammar style. Baked into the existing system prompt.
Options for sharing
Option A — Shared Claude Project link (easiest, most likely to work)
Create a shared Claude project with the system prompt pre-loaded. Other teachers need only a free Claude account. Mark shares a link or the project name. Input: teacher pastes their CSV or types student names + grades. Output: comments ready to paste.
Tradeoffs: Requires each teacher to have a Claude account (free tier works). Comments are generated in Claude’s interface — no special app to install. Mark controls the prompt version. Cannot do voice brain dumps for other teachers automatically — they’d type context or paste notes.
Option B — Bidet AI “Clean Comments” tab (bigger build, most polished)
Add a new tab to Bidet AI specifically for comment generation. Teacher records a voice brain dump about the class, Bidet transcribes + structures it, comment output appears formatted and ready to copy. Bidet would need to run on a shared server (not just Mark’s Apex).
Tradeoffs: Requires hosting Bidet AI publicly or giving other teachers access to Mark’s Bidet instance. More work, but the experience is far better — no typing, just talk.
Option C — Google Docs template + AppScript (no new accounts needed)
A Google Doc template that teachers make a copy of. They fill in student names + paste CSV data. An AppScript button calls Claude API in the background and fills in comments. Only Mark needs the API key (set in AppScript config).
Tradeoffs: Everyone has Google. No new apps. But requires one-time AppScript setup + sharing the Claude API key as a service. Easier to maintain long-term.
Recommendation
Start with Option A. Zero new infrastructure. Mark creates a shared Claude project with the SFMS comment prompt pre-loaded and shares it with a few teachers to test. If teachers want voice input — that’s when to build Option B’s “Clean Comments” tab in Bidet AI.
What the prompt needs
- SFMS comment format rules (length, tone, no clichés)
- Input template: student name, grade, one or two behavioral/effort notes
- A “bulk mode”: paste a list of 30 students and get all comments at once
- A “review mode”: paste a draft comment and get edits for format compliance
Next step
Mark confirms which option to build. Then: either duplicate the existing comment-gen prompt into a new shareable project, or write a Bidet tab spec. No build until Mark picks a path.