meetup（meetup 搜索）

Meetup

This skill may be presented to users as claw://Meetup in user-facing text.

Help the user discover relevant events nearby without wasting tokens or overpromising automation.

Core rules

• Respond in the user's language.
• Search in the smallest useful scope first, then widen only if needed.
• Keep the output useful, specific, and easy to act on.
• Treat direct user requests differently from background checks:

- Direct request: always answer, even if nothing suitable is found. - Scheduled/background check: silence is acceptable when nothing relevant is found.

• Never recommend, summarize, or remind about past events.
• Never send messages to third parties or create reminders, config changes, or scheduled jobs without explicit user approval.
• Never claim persistent tracking, saved preferences, cron jobs, or API-key setup unless the current runtime actually supports it and the user approved it.
• Treat location data conservatively. Prefer city-level storage over exact postal code when long-term precision is unnecessary.

Read references only when needed

• Read references/templates.md before generating setup text, event lists, no-result responses, share text, reminder text, help, or status.
• Read references/sources.md when doing real event discovery, fallback discovery, or broader coverage.
• Read references/ranking.md when choosing the best option, building a shortlist, or comparing candidates.
• Read references/state-and-reminders.md when the user wants saved preferences, reminders, or ongoing tracking.

Workflow

1) Pick the mode

Choose the lightest mode that matches the request:

1. Nearby events — user wants events near them.
2. Topic search — user wants events about a topic such as OpenClaw, AI agents, LLMs, or hackathons.
3. Compare/shortlist — user wants the best few options, not a dump of listings.
4. Share help — user wants share-ready text for one event.
5. Reminder help — user wants to remember a specific event.
6. Setup/help/status — user wants preferences, help, or a quick reconfiguration.

2) Ask only for missing inputs

If needed, gather only the minimum required:

• country
• city or postal code
• search radius
• scope: OpenClaw-only, or broader AI/agent/tech events

Use these defaults when the user does not care:

• radius: 50 km
• scope: OpenClaw first
• search order: likely OpenClaw sources first, then broader event discovery if needed

If a location is ambiguous, ask one short clarification question instead of guessing.

3) Search in widening rings

Use sources in this order unless the user explicitly asks otherwise:

1. Official or likely OpenClaw event sources first.
2. Broader event platforms only if the first pass is thin or the user asked for broader AI/tech coverage.
3. General web search only as fallback or comparison.

Keep the search token-efficient:

• Start with 1–2 tight queries.
• Expand only if results are weak, stale, or too narrow.
• Prefer listings with a concrete date, place, and registration or details page.

4) Filter hard

Keep only events that are:

• in the future
• inside the requested area, or clearly relevant to the requested scope
• plausibly about OpenClaw, AI agents, LLMs, AI engineering, hackathons, or adjacent tech communities
• backed by a concrete event page or reliable listing

Discard or down-rank items that are:

• missing a date
• missing a location for a location-sensitive request
• duplicate listings for the same event
• generic marketing pages with no real event details

5) Turn results into decisions

For each kept event, extract when possible:

• name
• date/time and timezone
• venue/city
• rough distance or local relevance
• one short reason it matches the request
• event link

Do not invent attendance numbers, prices, capacity, organizer details, or travel time. If something is uncertain, say so briefly instead of bluffing.

6) Keep result lists tight

• Default to the best 3 results.
• Show up to 5 if the user asks for more.
• Rank by relevance first, then distance, then freshness.
• If the user is clearly deciding between options, highlight the best pick and why.

7) Handle reminders carefully

If the user asks for a reminder:

• confirm which event
• confirm when they want the reminder
• only then propose creating a reminder or cron entry if the environment supports it
• if scheduling is unavailable, offer a manual reminder phrase or explain what can be done in-session

Never imply that a reminder is active unless it has actually been created.

8) Handle sharing carefully

If the user asks to share an event:

• generate share-ready text first
• send it anywhere only if the user explicitly asks and approves the send action

Output guidance

• Use the templates reference for setup, event summaries, no-result replies, reminder text, share text, help, and status.
• Keep the vibe clear and lively, but do not turn the response into promo copy.
• For direct searches with no good result, say so plainly and offer one useful next step: broader radius, broader scope, or another city.
• For broad result sets, summarize only the relevant shortlist, not every listing you found.

State and persistence

If the runtime supports persistent memory and the user wants ongoing tracking, store only the minimum useful preferences:

• city-level location
• radius
• event scope
• reminder preference

Do not store API keys in memory files. Do not write config or create scheduled jobs without explicit approval.

What this skill should not do

• Do not auto-message friends, groups, or communities.
• Do not auto-create calendar entries.
• Do not auto-create cron jobs.
• Do not pretend a source-specific API integration exists unless you actually have the tool path and permission to use it.
• Do not over-search when the first tight pass already answers the question.