# Accretive Thinking agent guide

Accretive Thinking is a local practice resource for investment banking and consulting interviews. Start with `llms.txt`; `llms-full.txt` consolidates the agent documentation. No hosted deployment URL is defined.

## Use boundaries

- Treat candidate answers and exported practice history as private. Do not transmit them or make external calls unless the user separately asks.
- Numeric checks are narrow calculation feedback, not an interview score. Do not infer missing values or grade free-text reasoning as though the checker did.
- Do not claim that a candidate is ready for recruiting, interviews, or a job. Explanations and judgment need human review.
- The question bank identifies its scenarios as original fictional material based on standard concepts, not copied from course questions. It does not claim these were asked by specific firms.
- Voice is optional and separate from the local written loop. It requires server-side `OPENAI_API_KEY` and makes a provider request. Never put a key in browser code or documentation.

## Main files

- `index.html`: Accretive Thinking home page.
- `bank.html`: question-bank practice; `bank.js` and `practice-engine.js` implement the interface and numeric comparisons.
- `consulting.html` with `app.js`: consulting case exercise.
- `decision-case.html`: decision case.
- `learn/`: static topic index, 8 topic pages, and 16 question pages.
- `question-bank.json`: question content, inputs, references, and review notes.
- `study-cli.mjs`: local `list`, `question`, and `check` commands; no network or API key required.
- `grade-answer.mjs`: numeric checker and function-tool definition.
- `server.mjs`: loopback-only local HTTP server and optional voice route.
- `docs/`: schema and usage reference.

When documentation and implementation disagree, use the source files above and update the docs.