Prompt Factory.doc

# `Meta_prompting.Prompt_factory`

Generate and refine *prompt packs* – self-contained multi-section strings that
capture every rule, constraint, and behavioural switch the assistant must
follow.  The factory is **pure** and therefore trivial to unit-test; it never
hits the network and never mutates global state.

> **Why does the project wrap prompts this way?**  Keeping the entire
> instruction set in a single, machine-parseable blob avoids coordination
> issues between the orchestration layer and the model.  It also allows
> the runner to diff two prompts as plain text.

---

## API at a glance

| Function | Use-case |
|----------|----------|
| `create_pack` | Build a brand-new prompt pack from scratch. |
| `iterate_pack` | Produce an *update pack* that describes the minimal changes required to move an existing prompt to the next revision. |

Supporting record types encode the required metadata:

* `create_params` – metadata for the *create* flow.
* `iterate_params` – metadata for the *iterate* flow.
* `eagerness` – how aggressively the assistant should advance before handing back control.

---

## `eagerness`