[Guide 9/10] AI Coding Assistant Prompts: Generate, Debug, Refactor and Test Code

# workdays.py
from collections.abc import Iterable
from datetime import date, timedelta


def business_days_between(
    start: date, end: date, holidays: Iterable[date] = ()
) -> int:
    """Count weekdays (Mon-Fri) from start to end, inclusive, minus holidays."""
    if start > end:
        raise ValueError("start must be on or before end")

    holiday_set = set(holidays)
    count = 0
    current = start
    while current <= end:
        if current.weekday() < 5 and current not in holiday_set:
            count += 1
        current += timedelta(days=1)
    return count


# test_workdays.py
from datetime import date

import pytest

from workdays import business_days_between


def test_full_week_counts_five_days():
    assert business_days_between(date(2026, 10, 19), date(2026, 10, 25)) == 5


def test_weekday_holiday_is_skipped():
    culture_day = date(2026, 11, 3)  # Tuesday
    assert business_days_between(date(2026, 11, 2), date(2026, 11, 8), [culture_day]) == 4


def test_start_after_end_raises():
    with pytest.raises(ValueError):
        business_days_between(date(2026, 10, 25), date(2026, 10, 19))

# (trimmed: same-day, weekend-only and weekend-holiday tests)

Sample output: wording varies between runs and models

What it does: groups an array into an object keyed by the key function,
for example support tickets grouped by status.

Line by line:
- <T, K extends PropertyKey>: generics. K is limited to the types allowed
  as object keys (string | number | symbol).
- items.reduce(..., {} as Record<K, T[]>): folds every item into an
  empty "key -> items" object.
- (acc[key(item)] ||= []): logical OR assignment (ES2021). Creates the
  group's array if it does not exist yet, then returns it.
- .push(item); return acc: adds the item and passes the object on.

Example:
groupBy(tickets, (t) => t.status)
// {"open":[{"id":1,...},{"id":3,...}],"closed":[{"id":2,...}]}

Sample output: wording varies between runs and models

1) Root cause: csv.DictReader returns every field as a string, so
   row["amount"] is "120.50", not a number. Python cannot add a str to
   the int total, hence the TypeError; int() fails because "120.50" is
   not an integer literal.

2) Minimal fix: parse the amount as Decimal, which keeps cents exact.

   from decimal import Decimal
   ...
   total = Decimal("0")
   ...
           total += Decimal(row["amount"])

   Result: 210.40 (float would drift on other inputs: 0.1 + 0.2 is
   0.30000000000000004, wrong for money)

3) Other risk: row["status"] == "paid" silently skips "Paid " (case and
   whitespace). Consider row["status"].strip().lower().

Sample output: wording varies between runs and models

Complexity: the original compares every pair, O(n²) string comparisons,
plus a linear "not in duplicates" list search. The version below is O(n).

from collections import Counter


def find_duplicate_emails(emails: list[str]) -> list[str]:
    """Return lowercased emails that appear more than once, in first-seen order."""
    counts = Counter(email.lower() for email in emails)
    seen: set[str] = set()
    result: list[str] = []
    for email in emails:
        key = email.lower()
        if counts[key] > 1 and key not in seen:
            seen.add(key)
            result.append(key)
    return result

# bench_dupes.py (trimmed); the original is kept as find_duplicate_emails_old
emails = [f"user{random.randint(0, 3000)}@Example.com" for _ in range(5000)]
for case in (emails, [], ["[email protected]"], ["[email protected]", "[email protected]", "[email protected]", "[email protected]"]):
    assert find_duplicate_emails(case) == find_duplicate_emails_old(case)

old = timeit.timeit(lambda: find_duplicate_emails_old(emails), number=1)
new = timeit.timeit(lambda: find_duplicate_emails(emails), number=1)
print(f"old: {old:.3f}s  new: {new:.4f}s  speed-up: {old / new:.0f}x")

Sample output: wording varies between runs and models

import math


def js_round(x: float) -> int:
    """Match JavaScript Math.round: halves round up (toward +infinity)."""
    return math.floor(x + 0.5)


def line_total(unit_price_cents: int, qty: int, discount_pct: float) -> int:
    gross = unit_price_cents * qty
    discount = js_round(gross * discount_pct / 100)
    return gross - discount

Semantic differences handled:
1. Rounding: Math.round rounds halves up (2.5 -> 3); Python's round()
   rounds to even (2.5 -> 2), so js_round() reproduces JS.
2. Number size: JS integers are exact only to 2^53 - 1; Python ints are
   unbounded. Same results for realistic prices.

Test cases (unit_price_cents, qty, discount_pct -> expected):
- (125, 1, 2)   -> 122   naive round() gives 123
- (1999, 3, 10) -> 5397
- (350, 1, 5)   -> 332

Sample output: wording varies between runs and models

Plan
1. src/models/invoice.ts: add currency: string to the Invoice type and
   default it to "VND" in createInvoice().
2. src/routes/invoices.ts: extend the Zod schema with
   currency: z.string().regex(/^[A-Z]{3}$/).default("VND").
3. src/routes/invoices.test.ts: test that the default is applied, that
   "USD" is accepted, and that "vnd" and "VNDX" return 400.

Missing context: invoices.ts imports saveInvoice from src/db/invoices.ts,
which builds the INSERT, and the table needs a migration. Please share
src/db/invoices.ts and your migration layout before I write the diffs.

Sample output: wording varies between runs and models

Behaviors: masks all but the first local character; keeps a 1-char local
part; lowercases the domain; trims spaces; rejects malformed addresses.

# test_masking.py
import pytest

from masking import mask_email


@pytest.mark.parametrize(
    ("email", "expected"),
    [
        ("[email protected]", "j*******@example.com"),
        ("[email protected]", "[email protected]"),            # one-character local part
        ("[email protected]", "J***@example.com"),      # domain lowercased
        # (trimmed: whitespace case)
    ],
)
def test_masks_valid_addresses(email, expected):
    assert mask_email(email) == expected


@pytest.mark.parametrize(
    "bad",
    ["", "no-at-sign", "@example.com", "john@", "a@[email protected]"],
)
def test_rejects_invalid_addresses(bad):
    with pytest.raises(ValueError):
        mask_email(bad)

Sample output: wording varies between runs and models

| Severity | Line | Problem | Fix |
|----------|------|---------|-----|
| Critical | 5 | SQL injection: email is formatted into the query, so ' OR '1'='1 returns users | Parameterize: conn.execute("SELECT ... WHERE email = ?", (email,)) |
| High | 5-6 | SELECT * returns every column, including any password hash, to the client | Select only public columns (id, name, email) |
| High | 6 | dict(row) fails: sqlite3 rows are tuples unless row_factory is set | Set conn.row_factory = sqlite3.Row |
| Medium | 6 | No match returns None, so dict(None) raises and the API answers 500 | Return 404 when row is None |
| Medium | 3 | Missing email parameter queries for 'None' | Return 400 when email is absent |
| Low | 4 | Connection is never closed | Use with closing(sqlite3.connect(...)) |

Performance: none found.

Sample output: wording varies between runs and models

Pattern (use with re.fullmatch):
[0-9]{4}-(0[1-9]|1[0-2])-(0[1-9]|[12][0-9]|3[01])

- [0-9]{4}            four-digit year
- (0[1-9]|1[0-2])     month 01-12
- (0[1-9]|[12][0-9]|3[01])   day 01-31

Matches:  2026-10-20, 1999-01-01, 2024-02-29, 2026-12-31
Rejects:  2026-13-01, 2026-00-10, 2026-1-5, 2026-10-32

Not checked:
- Days per month: 2026-02-30 passes. Use date.fromisoformat() for that.
- I used [0-9], not \d: in Python, \d also matches non-ASCII digits
  such as full-width ２０２６.
- Use re.fullmatch rather than ^...$ with re.match: $ also matches
  before a trailing newline, so "2026-10-20\n" would pass.

Sample output: wording varies between runs and models

find /var/log -type f -name '*.log' -size +100M -mtime -7 -exec ls -lh {} +

-type f            regular files only
-name '*.log'      quoted so the shell does not expand the *
-size +100M        larger than 100 MiB (works in GNU and BSD find)
-mtime -7          modified less than 7 days ago
-exec ls -lh {} +  print sizes, batching many files into one ls call

Sample output: wording varies between runs and models

Done.
- route.ts: rateLimit(ip, { limit: 5, window: "10 m" }) before
  validation; 429 with { success: false, error } when exceeded.
- route.test.ts: 6 requests from one IP -> [200 x5, 429].
Checks: typecheck 0 errors, lint 0 warnings, vitest 48 passed (1 new).

Sample output: wording varies between runs and models

1. Assumptions: UTF-8 CSV with a header; "." as decimal separator.
2. To verify: csv.DictReader, decimal.Decimal (stdlib, Python 3.12).
3. Likely to break: "1,200.50", "Paid " with a space, an empty amount.

Sample output: wording varies between runs and models