# Local Discovery > local-discovery — Find local events, venues, and activities — ad-hoc web discovery when the user asks 'what's happening' or 'what should I do this weekend'. - **Type:** Skill - **Install:** `agentstack add skill-atlasomnia-hermes-custom-pack-local-discovery` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [AtlasOmnia](https://agentstack.voostack.com/s/atlasomnia) - **Installs:** 0 - **Category:** [Productivity](https://agentstack.voostack.com/c/productivity) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [AtlasOmnia](https://github.com/AtlasOmnia) - **Source:** https://github.com/AtlasOmnia/hermes-custom-pack/tree/main/skills/local-discovery ## Install ```sh agentstack add skill-atlasomnia-hermes-custom-pack-local-discovery ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Local & National Discovery Find events, activities, venues, and things happening in the user's area (default: the user's local area) or anywhere nationally. Covers two subdomains: 1. **Events** — concerts, festivals, comedy, things to do tonight/this weekend 2. **Venues** — bars, lounges, restaurants, nightlife spots matching specific criteria (vibe, amenities, atmosphere) ## When to Use - User asks about local events, things to do tonight/this weekend - "What's happening around here?" or similar discovery requests - User asks for venue recommendations with specific criteria (e.g., "cigar lounge," "speakeasy," "classy bar") — local OR national - User asks "find me X somewhere" without specifying a city — treat as national search - Need to surface relevant activities based on user interests ## Venue Discovery Workflow When the user asks for venues rather than events, use this approach: ### 1. Start with Google Maps for venue discovery (PRIMARY TOOL) Navigate directly to Google Maps search: ``` browser_navigate(url="https://www.google.com/maps/search/cigar+lounge+near++") ``` **Google Maps works reliably** even when Google Search CAPTCHAs. Returns listings with ratings, review counts, addresses, hours, phone numbers, and user-submitted photos. Click individual venues for detailed info, reviews, and photos (use "Vibe" photo filter for atmosphere shots). ### 2. Fall back to `web_search` if needed Search using specific criteria keywords (e.g., `"cigar lounge"`, `"speakeasy bar"`). Collect names, addresses, and phone numbers from search results. **PITFALL:** `web_search` relies on `ddgs` which is installed in the system Python (`~/Library/Python/3.9/...`) but Hermes runs from a venv that doesn't see it. If you get `"ddgs package is not installed"`, skip to step 3 immediately — don't retry or try to reinstall. This is a persistent env path issue, not transient. ### 3. Skip review aggregator sites for browser navigation — they CAPTCHA aggressively **PITFALL:** Yelp, TripAdvisor, DuckDuckGo, and Bing ALL serve CAPTCHAs to browser sessions. Don't waste time trying to extract reviews or listings from them via `browser_navigate` — go straight to venue websites. **EXCEPTION:** `web_extract` works on TripAdvisor despite browser CAPTCHA. For national/regional venue discovery, use `web_extract(urls=[tripadvisor_url])` instead of browser navigation. ### 4. Go directly to venue websites via `browser_navigate` Venue websites are the most reliable source for accurate, current info: - **Hours** — always verify from the official site, never trust Google's cached hours - **Dress code** — many upscale venues post this explicitly - **Age restrictions** — check before recommending - **Contact info** — phone and email ### 5. Go beyond the algorithmic top-10 when the category is the city's identity **PITFALL:** When a city is FAMOUS for a venue category (e.g. Louisville + bourbon, Nashville + music venues, NOLA + jazz clubs, Austin + BBQ, etc.), presenting just a Yelp/TripAdvisor top-10 feels insultingly thin. The user knows the city is dense in that category and expects the FULL directory — all options, organized by neighborhood. **Recovery pattern:** - Run multiple `web_search` queries with specific venue names and cross-streets to surface listings the algorithm may have buried - Cross-reference Yelp snippets, TripAdvisor snippets, and niche directory sites (e.g., cigarlounges.co) from search results — even when `web_extract` fails on the full page, the search snippets carry review counts, ratings, and addresses - Search for niche directory/blog articles specific to that category (e.g., "complete list of cocktail lounges", "every jazz club in New Orleans") - Organize results by neighborhood/area — this adds massive value for the user planning a visit or crawl - Present the full count upfront ("28 across the metro") so the user knows the list is comprehensive, not truncated ### 6. When the user wants **bar-first / coed / date-night** rather than a niche enthusiast scene Do **not** keep feeding them classic category leaders if the venue photos or vibe read as male-dominated / hobbyist-only. Pivot the search intentionally: - Reframe the target from **"cigar lounge"** to **"restaurant or cocktail lounge with cigar patio/garden/menu"**. - Search local lifestyle/tourism sources for **date night**, **Restaurant Row**, **outdoor dining**, **nightlife**, **hotel lounges**, and **craft cocktails**. - Check Reddit (`r/`) for lived-experience notes like *quiet*, *older crowd*, *good for couples*, *great restaurants around there*, *people watching*, or *bar hop after*. - Distinguish three different classes clearly: 1. **Guaranteed cigar infrastructure** — official site explicitly mentions cigar lounge/garden/menu/patio. 2. **Bar-first with likely cigar compatibility** — local/tourism sources mention cigars, smoking patio, or cigar menu, but the venue is primarily a restaurant/bar. 3. **Great vibe but cigar certainty weak** — good coed/date-night energy, but hookah/smoking policy or cigar policy is not verified. - Be honest when a place is **hookah-forward** rather than cigar-verified. - Prefer options where the venue identity reads **mixed crowd / couples / date night** over enthusiast-heavy cigar rooms when the user is going with a partner. Useful source types for this pivot: - official venue sites - Visit / International Drive listings - Date Night Guide / local lifestyle blogs - old.reddit.com threads in `r/` when mainstream extractors do not support Reddit ### 7. Present results concisely Format each venue with: name, address, phone, hours, vibe description, and why it fits the user's criteria. Group by neighborhood/area when the list is large. End with a clear recommendation based on their stated preferences. ## Workflow ### 1. Try `web_search` first (but expect failure) ``` web_search(query="events tonight in my city") ``` **PITFALL:** `web_search` relies on `ddgs` which is installed in the system Python (`~/Library/Python/3.9/...`) but Hermes runs from a venv that doesn't see it. If you get `"ddgs package is not installed"`, skip to step 2 immediately — don't retry or try to reinstall. This is a persistent env path issue, not transient. ### 0b (special case: niche-interest + multi-city search) When the user asks for events around a specific interest (e.g., "cigar events," "car shows," "DJ night," "food festival") across multiple cities or statewide: - Run parallel `web_search` calls with structured queries, e.g.: - `"cigar" event "" ` - `"cigar" tasting ` - `"cigar" event June 20-22 2026` - Use `web_extract` on Eventbrite's city-specific pages: - `https://www.eventbrite.com/d/--//` - `https://www.eventbrite.com/d/--/cigar/` - etc. - Check niche vendors' event calendars (e.g., Cigars International, specialty lounges) — they regularly host tastings and live-music cigar nights that general aggregators miss. - Present results grouped by city, then date; include only events with concrete details (date/time/location). This avoids the "only the local metro" trap when user interest is statewide. ### 2. Fall back to `web_extract` on known event sources These are the most reliable source types for a metro area's events: - **The metro's major newspaper events page** — Weekly roundups published every Monday. Most reliable source. Extracts well via `web_extract`. - **The regional visitor bureau's events calendar** — Has an events calendar but often redirects or blocks bots. Use as secondary. ### 3. If `web_extract` fails or returns sparse content, use `browser_navigate` Navigate to the newspaper's events page directly. The page renders server-side so it loads without JS execution issues. ### 4. Filter and present results - Group by date (tonight / Saturday / Sunday) - Highlight free events prominently - Include location, time, price, and link - Give a brief recommendation based on what you know about the user's interests - Keep it concise — one section per day, bullet format ### 4b. Late-night follow-up searches after a main event When the user asks follow-ups like "anything after 10pm?" after fireworks, parades, festivals, or family events: - Treat it as a **post-event nightlife / after-party search**, not just another pass over official civic event calendars. - Search both general web and Eventbrite city/category pages, e.g. `site:eventbrite.com July 4 after party`, ` nightlife after fireworks`, and city-specific Eventbrite discovery URLs. - Verify individual listings before recommending them. Eventbrite search snippets often surface irrelevant out-of-area events; open/extract the event page and confirm **city, venue, date, start/end time, and age restriction**. - Be explicit if no late fireworks exist. Offer adjacent late options instead: bar crawls, waterfront bars, hotel/resort parties, clubs, live music, or festivals that continue after the fireworks. - For late-night results, include **end time** prominently; it matters more than start time for this intent. ## Known Event Sources (metro example) ### Eventbrite (Niche + Multi-City Events) - **Reliable for niche interests** (cigar tastings, car shows, themed nights, etc.) via city-specific search pages: - Example: `https://www.eventbrite.com/d/--//` - Example: `https://www.eventbrite.com/d/--//` - Use `web_extract(urls=[eventbrite_url])` — it extracts event listings cleanly. - Especially valuable when user interest spans multiple cities or is highly specific; general aggregators miss these events. ## Venue-Specific Sources ## Pitfalls - **Don't retry `web_search` after a ddgs failure** — it's an env path issue, not transient. Switch tools immediately. - **Many event sites are JS-heavy SPAs** (Eventbrite, Meetup) that return blank to the browser or 404 to extractors. Prefer the metro's major newspaper as primary source. - **Bot detection is common** on visitor/tourism sites. If blocked, move to the next source rather than fighting it. - **Don't over-research** — the user wants a quick scan, not an exhaustive database. 3–5 relevant items per day is enough. - **Yelp and TripAdvisor CAPTCHA aggressively** (DataDome). Skip them for venue research — go straight to official websites. - **Google Maps/Reviews often blocks browser sessions** with recaptcha. Use `web_search` for initial discovery, then navigate directly to venue sites. - **Hours change frequently** — always verify from the official website, never trust cached or third-party data. - **Social media login walls** — Instagram and Facebook require authentication to view any content (photos, posts, business page details). Skip entirely for venue research. - **DuckDuckGo also CAPTCHAs** — "Select all squares containing a duck" challenge after first search. Don't waste time trying multiple searches. - **Bing serves Cloudflare challenges** — same fate as Google Search. Use Google Maps instead. - **`web_extract` works on TripAdvisor despite browser CAPTCHA** — `web_extract(urls=["https://www.tripadvisor.com/Attractions-g191-Activities-c20-t101-United_States.html"])` successfully extracts venue lists with ratings, locations, and review snippets even when browser navigation is blocked. Use this for national/regional venue discovery. - **Magazine/lifestyle articles extract well** — Sites like Haute Living (`hauteliving.com`) produce curated venue lists that `web_extract` handles cleanly. Search DDG/Bing for article URLs, then extract via `web_extract`. - **Venue concept mismatch is real** — some venue concepts (e.g., cigar bars with themed adult entertainment staff) don't exist in certain markets or nationally. After thorough research, report honestly rather than stretching a recommendation that doesn't fit the criteria. ## Output Format User expects concise, direct results organized by date. No preamble beyond a one-line intro. Format: **TONIGHT (day, date)** - **Event Name** — Time, Location. Price. Brief note. **SATURDAY, ** - ... End with a short recommendation based on user interests. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [AtlasOmnia](https://github.com/AtlasOmnia) - **Source:** [AtlasOmnia/hermes-custom-pack](https://github.com/AtlasOmnia/hermes-custom-pack) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-atlasomnia-hermes-custom-pack-local-discovery - Seller: https://agentstack.voostack.com/s/atlasomnia - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.