Dynamic prompt is not updated for each `agent_model.run`

TL;DR

Dynamic prompt is not updated for each agent\_model.run What was going on Issue pydantic/pydantic-ai#839 (closed, 21 comments): Hi there, Follow up on my slack question. However inside a single run many steps can occur, with the ability to mutate the state retained in ctx.deps (through tool calls).

Steps

1. Dynamic prompt is not updated for each agent_model.run
2. What was going on

Issue pydantic/pydantic-ai#839 (closed, 21 comments): Hi there, Follow up on my slack question. The dynamic prompt is very sweet to provide up-to-date context to the LLM. However inside a single run many steps can occur, with the ability to mutate the state retained in ctx.deps (through tool calls). At the moment dynamic prompt is updated at the very begining of each run and then is static for the whole run. I would like to request an option to have more fine-grained update for dynamic prompt so it can refresh for each agent_model.request().

3. The fix

If you need the prompt refreshed between steps inside a single run, use @agent.instructions instead of a dynamic prompt. Instructions are re-evaluated between tool-call rounds (since PR #1503, v0.1.1), which covers the per-step refresh the issue asked for.

When to use

You are seeing this: The dynamic prompt is very sweet to provide up-to-date context to the LLM. Use this skill when you run into "Dynamic prompt is not updated for each agent_model.run".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

Versions mentioned in the source: #1503 v0.1.1. If you are on something much newer or older, the details may have shifted.

Why this happens

The original report does not dig into a root cause. It documents the symptom and the fix that resolved it.

In gh:pydantic/pydantic-ai#839 the issue was: Issue pydantic/pydantic-ai#839 (closed, 21 comments): Hi there, Follow up on my slack question. The dynamic prompt is very sweet to provide up-to-date context to the LLM. However inside a single run many steps can occur, with the ability to mutate the state retained in ctx.deps (through tool calls). At the moment dynamic prompt is updated at the very begining of each run and then is static for the whole run. I would like to request an option to have more fine-grained update for dynamic prompt so it can refresh for each agent_model.request(). The verified answer: If you need the prompt refreshed between steps inside a single run, use @agent.instructions instead of a dynamic prompt. Instructions are re-evaluated between tool-call rounds (since PR #1503, v0.1.1), which covers the per-step refresh the issue asked for. Source: https://github.com/pydantic/pydantic-ai/issues/839