[BUG][Windows] Bot gateway E2E browser check crashes on cp1252 when Playwright stderr contains box-drawing characters

[Dhivya-Bharathy]

[BUG][Windows] Bot gateway E2E browser check crashes on cp1252 when Playwright stderr contains box-drawing characters

Metadata

Field	Value
Repository	https://github.com/MervinPraison/PraisonAI
Labels	bug, windows, praisonai-bot, browser, developer-experience, e2e
Severity	Medium — browser validation path broken on default Windows terminals; masks real Playwright install error
Component	e2e-platform-audit/bot_gateway_e2e.py, Playwright subprocess capture
Affected platforms	Windows 10/11 with cp1252 stdout (default PowerShell)
Live audit result	11 PASS / 3 FAIL (docker env, browser x2)

---

Executive summary

The live bot gateway E2E harness (e2e-platform-audit/bot_gateway_e2e.py --live-only) validates praisonai-bot, MCP, code execution, sandbox, and browser control end-to-end. On Windows with default cp1252 console encoding, the browser control checks fail with a secondary UnicodeEncodeError on box-drawing character ║ (U+2551) instead of surfacing the actionable root cause: Playwright browser binaries are not installed (playwright install).

The failure chain is:

1. _playwright_headless_probe() subprocess returns Playwright's framed install hint (uses ║ box-drawing borders)
2. The probe check records this in Report.detail via str(exc)[:200]
3. Report.add() calls bare print() on the detail string (line 63)
4. cp1252 stdout cannot encode ║ → outer UnicodeEncodeError replaces the Playwright message
5. Developer sees 'charmap' codec can't encode character '\u2551' instead of playwright install

11 of 14 live checks pass (gateway, MCP, code, sandbox). Docker failure is environment-dependent (daemon not running). Browser failure is a double bug: missing Playwright browsers plus encoding-unsafe error reporting.

---

Environment

Field	Value
OS	Windows 10 (10.0.19045)
Shell	PowerShell (legacy conhost)
Python	3.13.2 (E2E runner); gateway subprocess may use 3.11
Console encoding	cp1252
OPENAI_API_KEY	Present (required for live tests)
PRAISONAI_AUTO_APPROVE	true
Playwright Python package	Installed
Playwright browser binaries	Not installed (playwright install not run)
Docker daemon	Not reachable (separate env failure)
Package versions (audit)	praisonai-bot 0.0.41, praisonai-browser 0.0.8

---

Step-by-step reproduction

# 1. Clone repo
git clone https://github.com/MervinPraison/PraisonAI.git
cd PraisonAI

# 2. Install packages (editable monorepo or pip install -e ".[all]")
pip install -e ".[all]"
pip install websockets pytest pytest-asyncio

# 3. Ensure Playwright browsers are NOT installed (default fresh env)
#    Skip: playwright install chromium

# 4. Set API key and auto-approve
$env:OPENAI_API_KEY = "<your-key>"
$env:PRAISONAI_AUTO_APPROVE = "true"

# 5. Do NOT force UTF-8 — reproduce cp1252
Remove-Item Env:PYTHONUTF8 -ErrorAction SilentlyContinue

# 6. Run live-only E2E
python e2e-platform-audit/bot_gateway_e2e.py --live-only

Expected behavior

• Browser checks FAIL with clear message: Please run: playwright install
• Other live checks pass (gateway, MCP, code, sandbox)
• E2E harness completes without crashing its own stdout

Actual behavior

• Gateway, MCP, code, sandbox checks PASS
• browser control (playwright headless) FAIL — detail truncated but contains playwright install and ║ chars (visible in JSON report)
• browser control via bot message FAIL — detail shows encoding error, not Playwright hint
• Summary: === Done: 3 failed === (docker + browser x2)

---

Full terminal evidence

Live E2E console output (sanitized paths)

ERROR:    Traceback (most recent call last):
  File "<venv>/site-packages/starlette/routing.py", line 645, in lifespan
    await receive()
  ...
asyncio.exceptions.CancelledError

=== PraisonAI Bot Gateway E2E ===
Repo: <repo-root>
Logs: <repo-root>/e2e-platform-audit/logs/bot-gateway-e2e

[OK] import praisonai_bot — 0.0.41
[OK] import praisonai_code — 0.0.56
[OK] import praisonai_browser — 0.0.8
[OK] import praisonai_mcp — 0.0.8
[OK] import praisonai_sandbox — 0.0.6
[OK] OPENAI_API_KEY — present
[FAIL] docker sandbox live execute — docker daemon not reachable
[OK] gateway startup (praisonai-bot) — 127.0.0.1:18765
[OK] mcp gateway /mcp endpoint — agent tool listed (1 tools)
[OK] mcp tools/call via bot gateway — response len=10
[OK] praisonai-code execute via bot message — marker found
[OK] sandbox exec via bot message — marker found
[FAIL] browser control via bot message — 'charmap' codec can't encode character '\u2551' in position 62: character maps to <undefined>

=== Done: 3 failed ===
Report: <repo-root>/e2e-platform-audit/logs/bot-gateway-e2e/report_20260727_130141.json

JSON report excerpt (shows masked Playwright message)

The browser control (playwright headless) check captured the real error before the outer handler failed:

{
  "name": "browser control (playwright headless)",
  "status": "FAIL",
  "detail": "updated.       \u2551\r\n\u2551 Please run the following command to download new browsers: \u2551\r\n\u2551                                                            \u2551\r\n\u2551     playwright install                              ",
  "duration_s": 10.3
},
{
  "name": "browser control via bot message",
  "status": "FAIL",
  "detail": "'charmap' codec can't encode character '\\u2551' in position 62: character maps to <undefined>",
  "duration_s": 10.3
}

The JSON file is UTF-8 and preserves the Playwright hint; the console loses it due to cp1252 print().

---

Interactive analysis

Symptom → diagnosis → fix flowchart

flowchart TD
    A[bot_gateway_e2e.py --live-only] --> B[browser control check]
    B --> C[_playwright_headless_probe subprocess]
    C --> D{Browsers installed?}
    D -->|No| E[Playwright stderr with U+2551 box chars]
    D -->|Yes| F[Browser probe PASS]
    E --> G[str exc captured in Report.detail]
    G --> H[Report.add calls print detail]
    H --> I{stdout encoding cp1252?}
    I -->|Yes| J[UnicodeEncodeError on U+2551]
    I -->|No UTF-8| K[Playwright hint visible in console]
    J --> L[Outer except: charmap error shown]
    L --> M[Developer misdiagnoses as encoding bug]
    K --> N[Developer runs playwright install]
    J --> O[Fix: safe_print + ASCII sanitize subprocess output]
    O --> P[Console shows playwright install hint]

Loading

Failure sequence diagram

sequenceDiagram
    participant E2E as bot_gateway_e2e.py
    participant PW as Playwright subprocess
    participant R as Report.add / print()
    participant Out as stdout cp1252

    E2E->>PW: chromium.launch headless probe
    PW-->>E2E: stderr with ║ framed "playwright install"
    E2E->>E2E: Check FAIL detail = str(exc)[:200]
    E2E->>R: add(FAIL, detail with ║)
    R->>Out: print(f"[FAIL] ... {detail}")
    Out-->>R: UnicodeEncodeError U+2551
    R-->>E2E: exception in outer browser block
    E2E->>R: add(FAIL, "charmap codec...")
    Note over E2E: Real fix (playwright install) hidden from console

Loading

---

Root cause analysis

Layer 1: Missing Playwright browsers (environment)

_playwright_headless_probe() at e2e-platform-audit/bot_gateway_e2e.py lines 385–408 runs:

proc = await asyncio.create_subprocess_exec(
    sys.executable, "-c", script,
    stdout=asyncio.subprocess.PIPE,
    stderr=asyncio.subprocess.PIPE,
    env=os.environ.copy(),
)
...
if proc.returncode != 0:
    err = (stderr or b"").decode(errors="replace").strip()
    raise RuntimeError(err[-400:] if err else f"exit {proc.returncode}")

When Chromium is not installed, Playwright emits a multi-line banner using box-drawing characters (║ U+2551). This is correctly captured in RuntimeError message.

Layer 2: Encoding-unsafe console reporting (bug)

Report.add() at lines 60–63:

def add(self, check: Check) -> None:
    self.checks.append(check)
    icon = {"PASS": "OK", "FAIL": "FAIL", "SKIP": "SKIP"}.get(check.status, "?")
    print(f"[{icon}] {check.name}" + (f" — {check.detail}" if check.detail else ""))

Bare print() encodes check.detail to cp1252 stdout. When detail contains ║, Python raises UnicodeEncodeError.

Layer 3: Outer exception handler masks root cause

Browser block lines 571–640:

except Exception as exc:
    report.add(Check("browser control via bot message", "FAIL", str(exc)[:200], time.time() - t0))

If report.add() itself raises (or a prior add in the same block fails), the outer handler records the encoding error, not the Playwright install hint.

Note on PYTHONUTF8 in harness

The script sets os.environ.setdefault("PYTHONUTF8", "1") at line 30, but this does not reliably reconfigure an already-initialized stdout on all Windows Python builds. Audit was run with cp1252-active console; failure reproduced.

---

Impact assessment

Check	Result	Notes
Package imports	PASS	All five packages
Gateway startup	PASS	127.0.0.1:18765
MCP /mcp endpoint	PASS	Agent tool listed
MCP tools/call	PASS	Live LLM round-trip
praisonai-code via WS	PASS	E2E_CODE_OK marker
Sandbox via WS	PASS	E2E_SANDBOX_OK marker
Docker sandbox	FAIL	Env: daemon not running
Browser Playwright probe	FAIL	Browsers not installed
Browser via bot message	FAIL	Encoding masks Playwright hint

Severity justification: Medium

• Blocks Windows developers from diagnosing browser setup
• Does not affect Linux/macOS CI with UTF-8
• Real fix for users is playwright install, but they cannot discover it from console output

---

Proposed fixes

Fix 1: Encoding-safe Report.add() (required)

Before (e2e-platform-audit/bot_gateway_e2e.py, lines 60–63):

def add(self, check: Check) -> None:
    self.checks.append(check)
    icon = {"PASS": "OK", "FAIL": "FAIL", "SKIP": "SKIP"}.get(check.status, "?")
    print(f"[{icon}] {check.name}" + (f" — {check.detail}" if check.detail else ""))

After:

def _safe_console(text: str) -> str:
    import sys
    enc = getattr(sys.stdout, "encoding", None) or "utf-8"
    try:
        text.encode(enc)
        return text
    except UnicodeEncodeError:
        return text.encode(enc, errors="replace").decode(enc)

def add(self, check: Check) -> None:
    self.checks.append(check)
    icon = {"PASS": "OK", "FAIL": "FAIL", "SKIP": "SKIP"}.get(check.status, "?")
    line = f"[{icon}] {check.name}"
    if check.detail:
        line += f" — {_safe_console(check.detail)}"
    print(line)

Fix 2: Sanitize subprocess stderr before storing in Check.detail

In _playwright_headless_probe() after decoding stderr:

import re

def _ascii_sanitize(text: str) -> str:
    # Strip box-drawing and other non-ASCII for console-safe reports
    return re.sub(r"[^\x20-\x7E\r\n\t]", "?", text)

err = _ascii_sanitize((stderr or b"").decode(errors="replace").strip())
if "playwright install" in err.lower():
    raise RuntimeError(
        "Playwright browsers not installed. Run: playwright install chromium"
    )
raise RuntimeError(err[-400:] if err else f"exit {proc.returncode}")

Fix 3: Pre-flight browser check with ASCII-only message

Add early check before live gateway tests:

def _check_playwright_browsers(report: Report) -> bool:
    try:
        asyncio.run(_playwright_headless_probe())
        return True
    except RuntimeError as exc:
        msg = str(exc)
        if "playwright install" in msg.lower() or "download" in msg.lower():
            report.add(Check(
                "playwright browsers installed",
                "FAIL",
                "Run: playwright install chromium",
            ))
        else:
            report.add(Check("playwright browsers installed", "FAIL", _safe_console(msg)[:200]))
        return False

Skip browser bot checks when pre-flight fails — avoids cascading encoding errors.

Fix 4: Wrap report.add in try/except (defensive)

def add(self, check: Check) -> None:
    self.checks.append(check)
    try:
        self._emit(check)
    except UnicodeEncodeError:
        check = Check(check.name, check.status, "<detail omitted: non-ASCII>", check.duration_s)
        self._emit(check)

---

Workarounds

Workaround 1: Install Playwright browsers (fixes root environment issue)

playwright install chromium
python e2e-platform-audit/bot_gateway_e2e.py --live-only

Workaround 2: Force UTF-8 console

chcp 65001
$env:PYTHONUTF8 = "1"
$env:PYTHONIOENCODING = "utf-8"
python e2e-platform-audit/bot_gateway_e2e.py --live-only

Workaround 3: Read JSON report instead of console

python e2e-platform-audit/bot_gateway_e2e.py --live-only
Get-Content e2e-platform-audit/logs/bot-gateway-e2e/report_*.json -Encoding utf8 |
  ConvertFrom-Json | Select-Object -ExpandProperty checks |
  Where-Object { $_.name -like "*browser*" }

The JSON report preserves the Playwright install message even when console fails.

---

Test plan

Manual (Windows cp1252, browsers not installed)

• E2E completes without uncaught exception
• Console shows playwright install or ASCII equivalent — not charmap codec
• JSON report unchanged (still UTF-8 full detail)

Manual (Windows cp1252, browsers installed)

• browser control (playwright headless) PASS
• browser bot agent via WS message PASS or fails with LLM-related reason only

Unit test for safe console

def test_report_add_cp1252_safe(monkeypatch, capsys):
    monkeypatch.setattr(sys.stdout, "encoding", "cp1252")
    report = Report()
    report.add(Check("browser", "FAIL", "Please run playwright install \u2551"))
    out = capsys.readouterr().out
    assert "UnicodeEncodeError" not in out
    assert "playwright install" in out.lower()

CI recommendation

- name: Bot gateway E2E live (Windows)
  if: runner.os == 'Windows'
  run: |
    playwright install chromium
    python e2e-platform-audit/bot_gateway_e2e.py --live-only
  env:
    OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
    PRAISONAI_AUTO_APPROVE: "true"

---

Acceptance criteria

• bot_gateway_e2e.py --live-only never raises UnicodeEncodeError from Report.add()
• When Playwright browsers missing, console shows actionable playwright install hint in ASCII
• JSON report continues to store full UTF-8 detail
• No regression on UTF-8 terminals (Linux/macOS/Windows Terminal)
• Documentation mentions playwright install chromium as prerequisite for browser E2E

---

Related files

Path	Role
e2e-platform-audit/bot_gateway_e2e.py	Primary fix location — Report.add, _playwright_headless_probe
src/praisonai-bot/praisonai_bot/cli/features/bots_cli.py	Browser tool wiring (BrowserBaseTool, line ~980)
src/praisonai-browser/	Browser package doctor / install docs
e2e-platform-audit/logs/bot-gateway-e2e/report_*.json	UTF-8 evidence preserved

---

References

• Playwright install docs: https://playwright.dev/python/docs/browsers
• Evidence log: e2e-platform-audit/logs/full-audit/bot-gateway-live.txt (sanitized in this issue)
• Live E2E run 2026-07-27: 11/14 checks PASS

[praisonai-triage-agent]

👋 Thanks for opening this issue, @Dhivya-Bharathy!

A maintainer will review this shortly. In the meantime:

• Make sure you've included steps to reproduce (for bugs)
• Check existing issues for duplicates
• Review the documentation for related guides

Bug and enhancement reports are auto-triaged with Claude. Questions and other issues: a maintainer can comment @claude to trigger analysis.

Routing: Agent-callable tools → PraisonAI-Tools; lifecycle plugins (tracing/hooks/guardrails) → PraisonAI-Plugins.

[praisonai-triage-agent]

Claude finished @Dhivya-Bharathy's task —— View job

---

Analysis: referenced file not present in repository

I investigated the reported bug but cannot implement the fix because the file it describes does not exist in this repository.

• Searched for e2e-platform-audit/bot_gateway_e2e.py — not found
• Searched all tracked files (git ls-files '*bot_gateway_e2e*') — no match
• Searched full git history across all branches — never committed
• Confirmed the e2e-platform-audit/ directory does not exist

Findings

The issue describes a UnicodeEncodeError on box-drawing character ║ (U+2551) in Report.add() and _playwright_headless_probe() within e2e-platform-audit/bot_gateway_e2e.py. That file, the e2e-platform-audit/ directory, and the referenced logs/bot-gateway-e2e/ reports are not part of this repository — neither in the working tree nor in any branch's git history.

The harness appears to be a local/private audit script used to run the live E2E audit (the JSON reports live under e2e-platform-audit/logs/... which is also absent). Because the buggy code isn't in the repo, there is no source line here to patch.

Why I'm not creating a PR

Per the repository guidelines (keep the SDK lightweight and powerful, reject scope creep, keep fixes minimal), I should not author a brand-new ~600-line E2E audit harness — inventing the Report class, Playwright probe, gateway/MCP/sandbox/browser checks — solely to host the described fix. That would be creating substantial new functionality with no existing consumer in this repo, not a minimal bug fix.

How to proceed

If the harness should live in this repo, please either:

1. Commit the existing bot_gateway_e2e.py (and the e2e-platform-audit/ scaffolding) to a branch, then re-trigger @claude — I'll apply the encoding-safe Report.add() and the playwright install hint surfacing directly to the real source; or
2. Point me to the correct path/branch if the file lives elsewhere.

For reference, the minimal fix the issue proposes is sound and low-risk:

• Encoding-safe console emit in Report.add() (text.encode(sys.stdout.encoding, errors="replace").decode(...) fallback).
• Detect playwright install in Playwright stderr and raise an ASCII-only RuntimeError("... Run: playwright install chromium") so the actionable hint survives a cp1252 console.
• Keep the JSON report writing full UTF-8 detail unchanged (no regression on Linux/macOS/UTF-8 terminals).

I can implement this immediately once the file is available in the repo.
• Branch: claude/issue-3424-20260727-0853

[MervinPraison]

@claude validate again

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job

---

Re-validation of issue #3424

• Check if e2e-platform-audit/bot_gateway_e2e.py exists in repo
• Apply encoding-safe fix if present — N/A, file absent
• Test & create PR — not applicable

Result: file still not present — no code to patch

I re-validated as requested. The target file and its directory are still absent from this repository:

• git ls-files '*bot_gateway_e2e*' → no match
• ls e2e-platform-audit/ → No such file or directory
• git log --all -- '*bot_gateway_e2e*' → never committed on any branch
• Repo-wide search for _playwright_headless_probe / e2e-platform-audit → 0 hits

The described Report.add() / _playwright_headless_probe() code lives in a local/private audit harness that has not been committed here, so there is no source line to fix. Authoring a brand-new ~600-line E2E harness just to host the fix would be scope creep and is out of line with the lightweight-SDK guidelines (AGENTS.md).

To proceed

Commit the existing e2e-platform-audit/bot_gateway_e2e.py (and scaffolding) to a branch and re-trigger @claude. I'll then apply the minimal, low-risk fix directly to the real source:

1. Encoding-safe console emit in Report.add() — fall back to text.encode(sys.stdout.encoding, errors="replace").decode(...) so cp1252 consoles never raise UnicodeEncodeError on ║ (U+2551).
2. Surface the actionable hint — detect playwright install in Playwright stderr and raise an ASCII-only RuntimeError("... Run: playwright install chromium").
3. No regression — JSON report keeps full UTF-8 detail; Linux/macOS/UTF-8 terminals unaffected.
   • Branch: claude/issue-3424-20260727-0951

[MervinPraison]

@claude review again and see if its worth creating pr. use muti agents to review

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job

---

Re-validation of issue #3424 (multi-agent review)

• Verify e2e-platform-audit/bot_gateway_e2e.py exists (working tree + all branches)
• Check git history across all branches for the file
• Inspect the REAL browser package for an analogous encoding-unsafe bug
• Assess whether a PR is worth creating

Verdict: Not worth creating a PR — target file still absent, no analogous real bug

I re-validated thoroughly as requested. (Note: the parallel sub-agent runner errored out, so I ran the multi-angle verification directly.)

1. The described buggy file does not exist here

• git ls-files | grep e2e-platform-audit → no match
• git ls-files '*bot_gateway_e2e*' → no match
• git log --all --diff-filter=A -- '*bot_gateway_e2e*' → never added on any of the ~500 branches
• Repo-wide grep for _playwright_headless_probe / class Report … def add … print → 0 hits
• e2e-platform-audit/ directory → does not exist

The Report.add() / _playwright_headless_probe() code lives in a local/private audit harness that has never been committed. There is no source line to patch.

2. The real browser package does NOT have the same bug
I checked the actual committed browser code that the issue's "Related files" point at:

• src/praisonai-browser/praisonai_browser/diagnostics.py — uses structured DiagnosticResult objects; errors are stored in details dicts (JSON-serialized), not bare print() of raw subprocess stderr. No cp1252 ║ hazard.
• src/praisonai-browser/praisonai_browser/launcher.py:240-244 — Chrome subprocess stdout/stderr go to subprocess.DEVNULL, so no box-drawing stderr is ever printed to the console.
• src/praisonai-bot/.../cli/features/bots_cli.py exists but contains no analogous encoding-unsafe console emit of Playwright stderr.

So the "double bug" (missing playwright install + cp1252 UnicodeEncodeError on U+2551) is confined to the uncommitted harness. There is no equivalent defect in this repository to fix.

3. Creating a PR would be scope creep
Per AGENTS.md (keep the SDK lightweight, minimal fixes, no new surface area), authoring a brand-new ~600-line E2E audit harness solely to host the proposed fix — inventing Report, the Playwright probe, and the gateway/MCP/sandbox/browser checks with no existing consumer — is exactly the kind of scope creep the guidelines reject. I will not create such a PR.

How to proceed

Commit the existing e2e-platform-audit/bot_gateway_e2e.py (and scaffolding) to a branch and re-trigger @claude. I'll then apply the minimal, low-risk fix directly to the real source:

1. Encoding-safe console emit in Report.add() — fall back to text.encode(sys.stdout.encoding, errors="replace").decode(...) so cp1252 consoles never raise UnicodeEncodeError on ║.
2. Surface the actionable hint — detect playwright install in stderr and raise an ASCII-only RuntimeError("... Run: playwright install chromium").
3. No regression — JSON report keeps full UTF-8 detail; Linux/macOS/UTF-8 terminals unaffected.

Alternatively, if the harness should never live in this repo, the fix belongs in that private tool and the issue can be closed as out-of-scope.

---

No PR created: the file to fix is not in this repository, and no analogous bug exists in the committed code.
• Branch: claude/issue-3424-20260731-0656