feat(community): add prayer-times ability

[mahsumaktas]

Note: This is a re-submission of #216 after addressing review feedback from @uzair401.
Original PR was closed by the reviewer — reopening was not possible without write access.

Summary

Voice-activated Islamic prayer times assistant with automatic background reminders.

• Skill: "When is the next prayer?", "What are today's prayer times?", specific prayer queries, location setup, calculation method selection
• Background Daemon: 5-minute advance reminder before each prayer + notification at prayer time
• API: Aladhan Prayer Times API — free, no API key required
• Calculation methods: ISNA, MWL, Diyanet (Turkey), Karachi, Makkah, Egypt

Changes Since #216

Fixed both issues raised by @uzair401:

1. ✅ Removed resume_normal_flow() from the daemon loop — daemons run indefinitely, this call is incorrect there
2. ✅ Fixed send_interrupt_signal(): now awaited with no arguments, followed by speak() for the actual message; _safe_interrupt is now async

Verified against the Background and Alarm templates.

[github-actions]

✅ Community PR Path Check — Passed

All changed files are inside the community/ folder. Looks good!

[github-actions]

🔀 Branch Merge Check

PR direction: community/prayer-times → dev

✅ Passed — community/prayer-times → dev is a valid merge direction

[github-actions]

✅ Ability Validation Passed

📋 Validating: community/prayer-times
  ✅ All checks passed!

[github-actions]

🔍 Lint Results

✅ __init__.py — Empty as expected

Files linted: community/prayer-times/background.py community/prayer-times/main.py

✅ Flake8 — Passed

✅ All checks passed!

[uzair401]

Hello @mahsumaktas, Thanks for the submission again, here are a few suggestions to implement to make the conversation feel more natural for the user:

1. SYSTEM_PROMPT has no instruction for spoken output quality — the "unknown" intent path falls through to _speak_next_prayer silently, and if the LLM ever returns a QUESTION: style response or markdown, it would go straight into speak(). Add to the prompt: "Respond ONLY with valid JSON. No extra text, no markdown, no bullet points, no questions."

2. SYSTEM_PROMPT intent examples read like API documentation, not how a US speaker actually talks. Replace the inline examples with natural phrasing guidance — "what time is fajr" instead of "Query a specific prayer", "what are my times today" instead of "Query all today's prayer times", "I'm in Chicago, Illinois" instead of "user mentions a city/country".

3. _format_all_times produces a comma-joined string of all 6 prayers read as one run-on sentence — roughly 40–50 words depending on city. "Today's prayer times for {data['city']}: {formatted}" would read as "Today's prayer times for Chicago: Fajr: 5:12, Sunrise: 6:44, Dhuhr: 12:30..." — far too long and hard to follow on a speaker with no screen. Split into two speak() calls: the first three prayers, pause, then the last three. Or offer "Which prayer do you want?" as a follow-up instead of reading all six unprompted.

4. f"Calculation method updated to {method}." — speaks the raw method number (e.g. "Calculation method updated to 2") which means nothing to the user. Map the number to a name before speaking: {2: "ISNA", 3="MWL", 4="Makkah", ...} and say "Calculation method set to ISNA."

5. "I need your location to get accurate prayer times. What city and country are you in?" in _setup — 18 words, slightly over target for a standard prompt. Tighten to: "What city and country are you in?" (7 words — the reason is obvious from context).

6. f"Location set to {city}, {country}. You can change this anytime." — second sentence adds no value on a voice device (there's no visible menu to change it from). Drop it: "Got it, location set to {city}."

7. "Sorry, I couldn't determine your location. Please try again." in _setup — 10 words, fine for length, but offers no path forward. The user doesn't know what to say differently. Change to: "Didn't catch that — try saying your city and country, like 'Dallas, United States'."

8. f"I couldn't find the time for {prayer}. Available prayers are: {', '.join(PRAYER_NAMES)}." — reading a comma-joined list of 6 prayer names aloud is awkward TTS and over 20 words. Change to: "I don't have a time for {prayer}. Try asking for Fajr, Dhuhr, or another prayer by name."

9. "All prayers for today have passed. Fajr will be the next prayer tomorrow." — 14 words, within limit, but "Fajr will be the next prayer tomorrow" is a slightly stilted construction. Change to: "All done for today. Fajr is next, tomorrow morning." (9 words, more natural).

10. "Sorry, something went wrong. Please try again." appears in both files — acceptable length but gives no path forward on a voice device. Change to: "Something went wrong. Try asking again."

11. f"{name} is in {mins} minute{'s' if mins != 1 else ''}. Time to prepare." in the background daemon — the second sentence "Time to prepare." is abrupt and slightly robotic. Change to: "{name} in {mins} minute{'s' if mins != 1 else ''} — time to get ready." Same length, more natural spoken cadence.

12. The _setup flow uses user_response() instead of run_io_loop() — on a smart speaker user_response() typically doesn't re-prompt if the user says nothing or the STT returns empty. Since location setup is a required gate, use run_io_loop("What city and country are you in?") so the device re-asks if it gets no response.

13. No hardcoded string matching issues found — intent routing correctly goes through the LLM classifier throughout. No exit/confirmation word lists exist in this ability. No menu-driven sequential prompts found. These areas are clean.

[mahsumaktas]

Thanks @uzair401 for the thorough voice UX review! All 12 suggestions implemented:

1. ✅ SYSTEM_PROMPT: Added 'Respond ONLY with valid JSON' — prevents markdown/questions leaking into speak()
2. ✅ Natural intent examples: Replaced API-style docs with spoken phrasing ('what time is fajr' instead of 'Query a specific prayer')
3. ✅ Split all_times: Now two speak() calls (Fajr/Sunrise/Dhuhr, then Asr/Maghrib/Isha) — no more 50-word run-on
4. ✅ Method name mapping: 'Calculation method set to ISNA' instead of '...to 2'
5. ✅ Setup prompt: Shortened to 'What city and country are you in?' using run_io_loop() for re-prompting
6. ✅ Dropped: 'You can change this anytime' — no visible menu on voice device
7. ✅ Location error: Added example hint ('try saying Dallas, United States')
8. ✅ Specific prayer error: Shortened, suggests asking by name
9. ✅ All passed message: 'All done for today. Fajr is next, tomorrow morning.'
10. ✅ Generic error: 'Something went wrong. Try asking again.'
11. ✅ Daemon reminder: '{name} in {mins} minutes — time to get ready.'
12. ✅ Setup uses run_io_loop(): Re-asks if STT returns empty

[uzair401]

Approved for community.
We will proceed with adopting this ability for a 30-day evaluation period to monitor performance, identify any bugs, and implement any necessary fixes or optimizations. If the ability proves stable and meets our quality standards during this period, we will move forward with publishing it to the marketplace, ensuring full credit is given to the original author for their work and contribution.
We truly appreciate your effort and encourage you to continue developing and submitting new abilities. Contributions like yours help strengthen the ecosystem, and we look forward to seeing more of your work.