Debugging Streamlit Apps
SkillDev toolsThis skill lets an agent run and debug Streamlit apps with hot-reload. It starts the Python backend and React frontend together using make debug, writes per-session logs, and lets the agent inspect those logs or take screenshots of the running app. Use it when testing code changes, investigating bugs, or checking UI behavior.
Use Debugging Streamlit Apps in Claude, ChatGPT or Ahel Desktop
Free. Sign in, add Debugging Streamlit Apps and connect your AI. About a minute.
Also: Claude Code · Cursor · Codex
Then ask your AI: use the Debugging Streamlit Apps skill
Details
Instructions available. Your AI can read the instructions. Execution depends on the setup they require.
Account requirements not reviewed. Check the skill instructions before use; ahel provides instructions and does not run this skill.
No other account needed.
Have a Streamlit app script ready to run.
What your AI can do with it
- Start a Streamlit app with make debug and hot-reload
- Inspect backend.log for Python prints, errors, and tracebacks
- Inspect frontend.log for console.log output and React errors
- Search logs with rg for specific patterns and context
- Write temporary Playwright scripts to interact with the app
- Capture screenshots of the running app
Getting started
- Have a Streamlit app script ready to run.
- Run make debug with the path to your app script.
- Note the app URL and the session log directory printed at startup.
- Search the backend and frontend logs with rg to find errors or debug output.
- Add print or console.log statements as needed, then rerun or rely on hot-reload.
What this skill tells your AI
The instructions your AI receives, as published by streamlit/streamlit in .claude/skills/debugging-streamlit/SKILL.md and read by ahel’s review.
Quick Start
make debug my_app.py
This starts both backend (Streamlit/Python) and frontend (Vite/React) with hot-reload. The app URL is printed on startup (default http://localhost:3001; 3000 is reserved for manual make frontend-dev; it may use 3002+ if other debug sessions are running). Avoid pinning VITE_PORT unless you have a specific hard requirement (last resort).
Hot-reload behavior:
- Frontend: Changes to
frontend/code are applied within seconds. - Backend: Only changes to the app script trigger a rerun. Changes to the Streamlit library itself (
lib/streamlit/) require restartingmake debug.
Log Files
Each make debug run writes logs to a per-session directory under work-tmp/debug/ and updates work-tmp/debug/latest/ to point at the most recent session.
Because latest/* is a symlink, it can move if multiple debug sessions are starting/stopping concurrently—prefer using the session directory path printed by make debug when you need stable log references.
You can find the exact session directory in the make debug startup output under the Log files section.
| File | Content |
|---|---|
work-tmp/debug/latest/backend.log | Python print() statements, Streamlit logs, errors |
work-tmp/debug/latest/frontend.log | Browser console.log(), React errors, Vite output |
Logs are cleared at the start of each session and persist after exit for post-mortem analysis.
Log size warning: Logs can grow large during extended debugging sessions. Instead of reading entire log files, use rg to search for specific patterns:
# Search for specific debug messages
rg "DEBUG:" work-tmp/debug/latest/backend.log
# Search for errors (case-insensitive)
rg -i "error|exception|traceback" work-tmp/debug/latest/backend.log
# Search with context (3 lines before/after)
rg -C 3 "my_function" work-tmp/debug/latest/backend.log
# Search frontend logs for specific component
rg "MyComponent" work-tmp/debug/latest/frontend.log
Use this directory for all debugging artifacts (scripts, screenshots, etc.) to keep them organized.
Adding Debug Output
Backend (Python):
print(f"DEBUG: session_state = {st.session_state}")
Frontend (TypeScript/React):
console.log("DEBUG: props =", props)
Frontend console.log() output appears in work-tmp/debug/latest/frontend.log (or the current session's frontend.log file).
Workflow
- Create or use a test script in
work-tmp/debug/(e.g.,work-tmp/debug/test_feature.py) - Run
make debug work-tmp/debug/test_feature.py - Verify startup: Check
work-tmp/debug/latest/backend.logforError/Exceptionandwork-tmp/debug/latest/frontend.logfor console errors to ensure both servers started correctly - Access the printed App URL in your browser (default
http://localhost:3001, but it may be3002+) - Verify script execution: Check
work-tmp/debug/latest/backend.logagain for any errors after the first app access - Monitor logs by inspecting
work-tmp/debug/latest/backend.logandwork-tmp/debug/latest/frontend.log - Edit code - changes apply automatically via hot-reload
- Check logs for debug output
Quick error check:
# Backend errors
rg -i "error|exception" work-tmp/debug/latest/backend.log
# Frontend console errors
rg -i "error" work-tmp/debug/latest/frontend.log
Temporary Playwright Scripts for Screenshots & Testing
For advanced debugging with screenshots or automated UI interaction.
Quick: Playwright CLI
For simple screenshots and interactions, use @playwright/cli (available in frontend devDependencies):
cd frontend
STREAMLIT_APP_URL=http://localhost:3001
yarn playwright-cli open "$STREAMLIT_APP_URL"
yarn playwright-cli screenshot --filename ../work-tmp/debug/screenshot.png --full-page
yarn playwright-cli close
See https://github.com/microsoft/playwright-cli for more commands (snapshot, click, fill, etc.).
Custom Scripts
For complex interactions, create temporary Playwright scripts in work-tmp/debug/:
# work-tmp/debug/debug_screenshot.py
"""Temporary Playwright script for debugging - run against make debug."""
import os
from playwright.sync_api import sync_playwright, expect
from e2e_playwright.shared.app_utils import get_text_input, click_button
from e2e_playwright.conftest import wait_for_app_loaded, wait_for_app_run
def main():
app_url = os.environ.get("STREAMLIT_APP_URL", "http://localhost:3001")
with sync_playwright() as p:
browser = p.chromium.launch(headless=True)
page = browser.new_page(viewport={"width": 1280, "height": 720})
# Connect to app started with `make debug`
page.goto(app_url)
wait_for_app_loaded(page)
# Interact with the app
text_input = get_text_input(page, "Name")
text_input.fill("Test User")
click_button(page, "Submit")
wait_for_app_run(page)
# Verify and screenshot
expect(page.get_by_text("Hello, Test User")).to_be_visible()
page.screenshot(path="work-tmp/debug/debug_screenshot.png", full_page=True)
print("Screenshot saved to work-tmp/debug/debug_screenshot.png")
browser.close()
if __name__ == "__main__":
main()
Running Temporary Scripts
Ensure make debug <app.py> is running first (start it in a background task if needed). If your make debug session is using a non-default port, set STREAMLIT_APP_URL accordingly, then run the Playwright script:
STREAMLIT_APP_URL=http://localhost:3001 \
PYTHONPATH=. uv run python work-tmp/debug/debug_screenshot.py
This uses the uv-managed environment with all dependencies (playwright, etc.) and makes e2e_playwright importable without path manipulation.
Available Utilities from e2e_playwright
Element Locators & Interactions (e2e_playwright.shared.app_utils):
Provides helpers like get_text_input(), get_button(), click_button(), get_checkbox(), etc.
Synchronization (e2e_playwright.conftest):
wait_for_app_loaded(page)- wait for initial loadwait_for_app_run(page)- wait for script execution after interactionwait_until(page, fn, timeout)- poll until condition is true
Playwright API Reference: https://playwright.dev/python/docs/api/class-playwright
Screenshot Best Practices
# Full page screenshot
page.screenshot(path="work-tmp/debug/full.png", full_page=True)
# Element screenshot
element = page.get_by_test_id("stDataFrame")
element.screenshot(path="work-tmp/debug/dataframe.png")
Troubleshooting
Port already in use / multiple sessions:
make debugwill automatically pick a free frontend port (typically in the3001-3100range) so multiple debug sessions can run simultaneously.- Frontend port
3000is reserved for manualmake frontend-devsessions. - If you have a hard requirement for a specific frontend port, you can pin it with
VITE_PORT=3002 make debug <app.py>(last resort).
Hot-reload not working:
- Backend: Only the app script is watched. Changes to
lib/streamlit/require restartingmake debug. - Frontend: Check
work-tmp/debug/latest/frontend.logfor Vite errors. TypeScript errors can break HMR.
Playwright script fails to connect:
- Verify
make debugis running and healthy - Check the printed App URL is accessible in the browser
- Ensure
wait_for_app_loaded(page)is called afterpage.goto()
Cleanup
After debugging is complete, remove temporary scripts and screenshots from work-tmp/debug/.
Related Skills
- understanding-streamlit-architecture skill: For deeper understanding of backend/frontend internals when debugging cross-layer issues
Signals
- GitHub stars
- 46k
- Forks
- 4k
- Last commit
- Oct 2026
Others that do the same job
Questions
- Does it support hot-reload?
- Yes. Frontend changes under frontend/ apply within seconds. Backend changes only trigger a rerun for the app script; changes to the Streamlit library itself require restarting make debug.
- Where are the logs written?
- Each make debug run writes logs to a per-session directory under work-tmp/debug/ and updates work-tmp/debug/latest/ to point at the most recent session.
- Can it take screenshots of the running app?
- Yes. The agent can write temporary Playwright scripts to interact with the app or capture screenshots.
- What port does the app use?
- The app URL is printed on startup. The default is http://localhost:3001; 3000 is reserved for manual make frontend-dev, and it may use 3002+ if other debug sessions are running.
Advanced
- Item type
- skill
- Key
debugging-streamlit-streamlit- Source
- github.com/streamlit/streamlit
github.com/streamlit/streamlit
Related picks
Skill · wshobson
The pick for Pythonpython-pro
Skill · jeffallan
The pick for Pythonsetup-ts-deep-modules
Skill · mattpocock
The pick for TypeScripttypescript-pro
Skill · jeffallan
The pick for TypeScriptreact-component-performance
Skill · davila7
The pick for Reactreact-doctor
Skill · millionco
The pick for React