Debug Prompt

Auto‑generates AI‑friendly debug prompts for failed Vedro scenarios, helping you quickly spot root causes and apply the smallest possible fix.

✨ Features

• Zero‑config. Enable the plugin and you’re done.
• Automatic prompt generation. On every failed scenario a Markdown file is produced with steps, source, error message & traceback.
• LLM‑ready. A built‑in system prompt instructs ChatGPT (or any LLM) to analyse, locate the bug and suggest a minimal patch.

🛠 Installation

Quick

For a one‑liner install via Vedro’s plugin manager:

$ vedro plugin install vedro-debug-prompt

Manual

Prefer manual steps? No problem:

1. Install the package:

$ pip install vedro-debug-prompt

2. Enable the plugin in vedro.cfg.py:

import vedro
import vedro_debug_prompt

class Config(vedro.Config):

    class Plugins(vedro.Config.Plugins):

        class DebugPrompt(vedro_debug_prompt.DebugPrompt):
            enabled = True

🚀 Usage

Run your tests as usual:

$ vedro run

Example output when a scenario fails:

Scenarios
* 
 ✗ decode base64 encoded string
   |> AI Debug Prompt: .vedro/tmp/prompt_liywiyo1.md
   ✔ given_encoded_string
   ✔ when_user_decodes_string
   ✗ then_it_should_return_decoded_string
╭───────────────────── Traceback (most recent call last) ──────────────────────╮
│ /scenarios/decode_base64_str.py:20 in then_it_should_return                  │
│                                                                              │
│   17         self.result = b64decode(self.encoded)                           │
│   18                                                                         │
│   19     def then_it_should_return_decoded_string(self):                     │
│ ❱ 20         assert self.result == "banana"                                  │
│   21                                                                         │
╰──────────────────────────────────────────────────────────────────────────────╯
AssertionError
>>> assert actual == expected
    - 'banana' 
    + b'banana' 
 
 
# --seed 280dc986-e618-4102-aa53-056c2876c00e
# 1 scenario, 0 passed, 1 failed, 0 skipped (0.00s)

Open the referenced prompt_*.md file and paste its contents into ChatGPT.

tip

In most terminals you can ⌘‑click (or Ctrl‑click on Linux/Windows) the file path to open it instantly

A typical response might look like:

1. Root cause The call to b64decode returns a bytes object (b'banana'), but the test asserts against a Python str ("banana"), so the types (and values) don’t match.

2. Suggested fix (code)
   Apply this minimal patch in scenarios/decode_base64_str.py:

 def when_user_decodes_string(self):
-    self.result = b64decode(self.encoded)
+    self.result = b64decode(self.encoded).decode('utf-8')

3. Why this works
   By calling .decode('utf-8') on the bytes result, you convert it to a str ("banana"), which matches the test’s expected value.

⚙️ Custom Prompts

Need a different tone, extra context, or even a completely new set of instructions for the LLM? You can swap out the built‑in PromptBuilder for your own implementation and override just the parts you care about. The most common tweak is changing the system prompt.

Example: overriding the system prompt

# vedro.cfg.py
import vedro
import vedro_debug_prompt


class MyPromptBuilder(vedro_debug_prompt.PromptBuilder):
    def _get_system_prompt(self) -> str:
        # Tell the LLM exactly how you want it to behave
        return "<system prompt>"


class Config(vedro.Config):
    class Plugins(vedro.Config.Plugins):
        class DebugPrompt(vedro_debug_prompt.DebugPrompt):
            enabled = True
            prompt_builder = MyPromptBuilder()  # ← use the customised builder

That’s it! The DebugPrompt plugin will now use your custom system prompt whenever it generates a prompt.