Skip to content

heymegabyte/agent-skills / 15-site-generation

site-generation

Unverified

End-to-end AI website generation pipeline. Claude Opus 4.8 emits Bolt-style <boltArtifact> envelopes (multi-file, plan-first) that customize Vite+React+Tailwind templates from pre-researched business data. Pre-research via APIs, media acquisition, brand extraction, visual inspection via GPT Image 2 vision, R2 upload (per-file content-type by extension), D1 status updates. Supports all business types: SaaS, portfolio, non-profit, restaurant, salon, medical, legal, retail, tech.

FileKindSize
SKILL.mdskill md11.7 KB
_real-ua.mjsscript1.4 KB
blog-import.mjsscript10.5 KB
bolt-artifact-protocol.mdmarkdown8.4 KB
build-breaking-rules.mdmarkdown86.0 KB
build-image-redirects.mjsscript3.4 KB
build-prompts.mdmarkdown86.3 KB
check-routes.shexecutablescript904 B
domain-features.mdmarkdown17.1 KB
enhance-blog-posts.mjsscript12.9 KB
homepage-block-library.mdmarkdown12.6 KB
local-seo.mdmarkdown7.6 KB
media-acquisition.mdmarkdown39.2 KB
non-technical-owner-onboarding.mdmarkdown8.3 KB
page-set-expansion.mdmarkdown19.1 KB
pseo-templates.mdmarkdown13.3 KB
quality-gates.mdmarkdown46.6 KB
research-pipeline.mdmarkdown20.5 KB
scan-assets.mjsscript3.0 KB
small-business-mode.mdmarkdown9.6 KB
source-fidelity-loop.mdmarkdown10.0 KB
squarespace-full-crawl.mjsscript2.8 KB
template-improvements-100.mdmarkdown15.0 KB
template-system.mdmarkdown39.7 KB
validate-assets.mjsscript5.0 KB
validate-banned-slop.mjsscript4.6 KB
validate-brand-colors.mjsexecutablescript6.1 KB
validate-citations.mjsscript6.0 KB
validate-color-contrast.mjsscript5.2 KB
validate-favicon-set.mjsscript4.9 KB
validate-html-entities.mjsscript3.4 KB
validate-image-relevance.mjsscript6.1 KB
validate-internal-links.mjsscript4.3 KB
validate-jsonld-schema.mjsscript5.1 KB
validate-nap-consistency.mjsexecutablescript8.0 KB
validate-page-count.mjsscript4.1 KB
validate-page-set-completeness.mjsscript9.3 KB
validate-photo-authenticity.mjsexecutablescript10.0 KB
validate-pwa-kit.mjsscript7.0 KB
validate-route-metadata.mjsscript7.3 KB
validate-source-fidelity.mjsscript6.0 KB
validate-typography.mjsscript4.0 KB
verticals/local-business-site.mdmarkdown3.8 KB
verticals/nonprofit-site.mdmarkdown3.3 KB
verticals/restaurant-site.mdmarkdown2.6 KB

research-pipeline.md

View on GitHub
---
name: "research-pipeline"
description: "API-driven business research: Google Places, website scraping, social verification, brand extraction, confidence scoring. All research runs on Worker before container."
updated: "2026-04-24"
---

# Research Pipeline

All research runs on the Worker, not in the container. Results written as `_`-prefixed JSON files into the build directory. Claude Code reads these — never calls APIs for research.

## Phase 0a: Business Profile (Google Places API)

**Query** — `GOOGLE_PLACES_API_KEY` → `findplacefromtext` with business name + address. Then `place/details` for:

- `name`, `formatted_address`, `formatted_phone_number`
- `opening_hours`, `website`, `rating`, `user_ratings_total`
- `reviews` (top 3)
- `photos` (download to R2)
- `geometry` (lat/lng), `types`, `price_level`, `business_status`

**Confidence**: `google_places` source at 80-95 depending on field.

### Fallback chain

1. **Yelp Fusion API** (`YELP_API_KEY`) — business match by name+location, returns reviews/photos/hours/categories (confidence 60-80)
2. **Facebook Graph API** — page search by business name, returns about/hours/phone/address (confidence 55-70)
3. **BBB API/scrape** — search by business name, returns rating/accreditation/complaints (confidence 70-85, trust signal)
4. **Workers AI (Llama 4 Scout) research prompt** — synthesize from web search results (confidence 50-70, LAST RESORT)

### Competitor analysis (auto, no extra API cost)

Google Places `nearbysearch` with same `type` within 5mi radius. Top 5 competitors: extract names, ratings, review counts, websites. Used in build prompt for differentiation ("You have 4.8 stars vs competitors averaging 4.2 — emphasize this").

## Phase 0b: Website Scraping (Deep Crawl)

If business has existing website:

- Crawl up to 50 pages
- For each: extract title, headings, body text, images (download to R2), nav structure, footer content, meta tags, schema.org data
- Store as `_scraped_content.json` keyed by URL path

**Extract**:

- All text content for reuse
- All image URLs for download
- Sitemap structure for page recreation
- Blog posts for content migration

### Sitemap fetch (***CRITICAL FOR URL PRESERVATION***)

- Before crawling, fetch `/sitemap.xml` and `/sitemap_index.xml`
- Parse all `<loc>` URLs into `_scraped_content.json.sitemap_urls[]`
- These become the canonical list of URLs that must resolve (200 or 301) on the new site
- If no sitemap exists, build URL list from crawled pages + any links discovered
- Store as `_original_urls: string[]` in `_scraped_content.json`

**Tools** — `fetch()` with CF Workers (no Puppeteer needed for static sites). For JS-rendered: use `@cloudflare/puppeteer` or skip with graceful degradation. Parse HTML with regex patterns (no DOM parser in Worker).

## Phase 0c: Social Media Verification

For each platform (Facebook, Instagram, Twitter/X, LinkedIn, YouTube, TikTok, Pinterest, Yelp, Google Business):

- Construct candidate URL from business name → HEAD request → verify 200 status
- Only include URLs at 90%+ confidence
- Dead links: exclude entirely

## Phase 0d: Brand Extraction (***PRIMARY COLOR RETRIEVAL***)

**Priority order for primary color**: logo dominant color → header/nav background → CTA button color → accent borders → hero overlay.

**Extract via**:

1. **Brandfetch API** (`BRANDFETCH_API_KEY`) — returns full brand kit (colors, fonts, logos) at 90% confidence
2. **Logo.dev API** (`LOGODEV_TOKEN`) — logo image → GPT Image 2 vision extracts dominant colors
3. **GPT Image 2 vision** on screenshot of existing site
4. **Color extraction** from downloaded images in `assets/` (look for signage, storefront, uniforms)

### Output to `_brand.json`

```json
{
  "colors": {
    "primary": { "value": "#8B1A2B", "source": "extracted_from_logo", "confidence": 0.92 },
    "secondary": { "value": "#1A1A2E", "source": "extracted_from_website", "confidence": 0.85 },
    "accent": { "value": "#E8B931", "source": "extracted_from_website", "confidence": 0.80 },
    "background": { "value": "#0D0D1A", "source": "derived_from_primary", "confidence": 0.88 },
    "foreground": { "value": "#F5F5F5", "source": "contrast_calculated", "confidence": 0.95 }
  },
  "fonts": { "heading": "Playfair Display", "body": "Inter", "source": "extracted_from_website" },
  "personality": "professional",
  "logo_url": "assets/logo.svg"
}
```

### Color source tracking (***CRITICAL***)

- Every color must have `color_source`: `extracted_from_logo|extracted_from_website|extracted_from_assets|derived_from_primary|contrast_calculated|generated`
- NEVER guess colors from business category
- The njsk.org burgundy incident: system guessed "warm soup kitchen colors" instead of extracting their actual burgundy brand
- `background` is derived by darkening primary by 80-90% lightness in OKLCH
- `foreground` is calculated for WCAG AA contrast against background

### New business fallback (no web presence)

If business has no website, no logo, and no social media:

1. **Google Places photos** → extract dominant color from storefront/signage images
2. **Google Street View screenshot** → extract from building facade, awning, signage
3. **Industry-neutral defaults** as LAST RESORT: primary=`#2563EB` (accessible blue), secondary=`#1E293B` (slate), accent=`#F59E0B` (amber)
   - Mark `color_source: "fallback_default"` with confidence 0.40
   - Flag in `_brand.json.warnings[]`: "No brand assets found — using neutral defaults. Business owner should provide logo/colors."
   - NEVER skip to category-based guessing

## Phase 0e: Confidence Scoring

Every data point gets `Conf<T>`:

```ts
{ value: T, confidence: number (0-1), sources: Source[], apa_citation: string, source_url: string, refId: string }
```

**Source types**: `google_places`, `llm_inference`, `user_provided`, `web_scrape`, `social_verify`, `peer_reviewed`, `gov_edu`, `primary_data`, `industry_research`

**Merge rule** — higher confidence wins, corroboration boosts +0.1 (capped at 0.99)

**UI policy**:

- Prominent **≥0.85**
- Standard **0.70-0.84**
- Deemphasize **0.50-0.69**
- Hide **<0.50**

### APA citation requirement (***NON-NEGOTIABLE — see rules/citations.md***)

Every quantitative field (%, N, $, ratio, comparison, year-claim) MUST carry `apa_citation` (APA 7th ed) and `source_url`. Examples:

- `apa_citation: "U.S. Bureau of Labor Statistics. (2024). Occupational employment statistics: Restaurant industry. https://www.bls.gov/oes/"`
- `apa_citation: "Brewer, S. (2024). AI search citation rates. Journal of Search Engine Optimization, 15(2), 88-104. https://doi.org/10.xxxx/xxxxx"`

**Thresholds**: Confidence ≥0.85 requires 2+ corroborating cites; single source = 0.70; unsourced = rejected.

### Warnings generated for missing

- phone (<0.5), email (<0.5), geo (<0.3), booking_url (<0.5), reviews (<0.3)
- apa_citation on any quantitative claim (auto-fail)

## Phase 0g: Citation Aggregation (`_citations.json`)

Sibling file to `_research.json`. APA 7th ed bibliography keyed by `refId`. Schema:

```json
{
  "ref-1": {
    "type": "journal|web|government|industry|primary",
    "authors": ["Brewer, S.", "Lee, M."],
    "year": 2024,
    "title": "AI search citation rates with structured data",
    "publication": "Journal of Search Engine Optimization",
    "volume": "15(2)",
    "pages": "88-104",
    "doi": "10.xxxx/xxxxx",
    "url": "https://...",
    "accessed": "2026-04-25",
    "apa_formatted": "Brewer, S., & Lee, M. (2024). AI search citation rates with structured data. Journal of Search Engine Optimization, 15(2), 88-104. https://doi.org/10.xxxx/xxxxx"
  }
}
```

- Container script `~/format-citations.js` (citation-js npm) converts BibTeX/RIS/CSL-JSON → APA 7th
- Write `_citations.json` during research phase
- Components reference by `refId`
- Build gate `validate-citations.js` ensures every numeric claim in dist/ HTML resolves to a `refId` entry

## Phase 0e2: Publication Deeplink Resolver

***FOR SITES WITH RESEARCH/PUBLICATION TILES — NEVER DUPLICATE-CONTENT STUB***

For every publication/paper/article/case-study item discovered in source crawl (academic CVs, faculty pages, "Selected Publications", `/portfolio` indexes, journal pages, ResearchGate scrapes), resolve a canonical external `deeplink_url` BEFORE rendering tile.

Internal `/portfolio/<slug>` stub pages that re-display only the citation block are a duplicate-content failure (lonemountainglobal.com 2026-05-01 incident: 12 publication tiles linked to internal stubs containing nothing but re-rendered citations, when actual article URLs existed).

**Build gate** — `validate-publication-deeplinks.mjs` (skill 15 quality-gates.md) fails on any publication tile without `deeplink_url` OR with `deeplink_url` pointing to an internal route.

### Discovery chain (try in order, stop on first 200-OK match)

1. **Source-site PDF/citation block** — pdftotext + LLM JSON extraction of `{title, authors, journal, year, doi?, pmid?, arxiv_id?}`
2. **Crossref API** — `https://api.crossref.org/works?query.bibliographic=<title>&query.author=<first-author>&rows=5` then match by year+author similarity → returns DOI → `https://doi.org/<doi>`
3. **PubMed E-utilities** — `https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term=<title>+AND+<author>[author]` → returns PMID → `https://pubmed.ncbi.nlm.nih.gov/<pmid>`
4. **arXiv API** — `http://export.arxiv.org/api/query?search_query=ti:<title>+AND+au:<author>` → returns arXiv ID → `https://arxiv.org/abs/<id>`
5. **Google Scholar via Serper** — `https://serper.dev/search?q=<title>+filetype:pdf` (rate-limit 1/sec) → top result if domain ∈ {jstor.org|sciencedirect.com|tandfonline.com|wiley.com|springer.com|nature.com|science.org|nih.gov|bmj.com|jamanetwork.com|nejm.org|pubs.acs.org|pubs.rsc.org|cambridge.org|oup.com|sage.com|elsevier.com|university journal hostnames}
6. **Tavily/Perplexity research API** — `<title> <author> <year>` → first authoritative URL
7. **Journal homepage search** if journal known

### Persistence

- Persist to `_research.json.publications[]` with `{ title, authors[], journal, year, doi?, pmid?, arxiv_id?, deeplink_url, deeplink_source, paraphrased_summary, image_url?, citation_apa }`
- `deeplink_source` ∈ `crossref|pubmed|arxiv|scholar|tavily|journal_search`
- **Confidence**: ≥0.85 on doi/pmid/arxiv hit, 0.70 on scholar/tavily, 0.50 on journal_search fallback
- NEVER mark `deeplink_url=null` and ship — fail build instead

### Render

- Tile wrapper: `<a href={deeplink_url} target="_blank" rel="noopener noreferrer" data-card="publication">` with entire surface clickable
- Inner action button "Read Full Paper →" points to SAME `deeplink_url` (validator compares card-wrapper href vs button href, fail on mismatch)
- Journal/conference name in tile body hyperlinked to journal homepage (institution-name anchor, NEVER "click here")
- Internal stub routes `/portfolio/<slug>` are FORBIDDEN for publication-typed cards — `_redirects` MUST emit `/portfolio/<slug> 301 <deeplink_url> 301` for any source URLs that previously pointed at internal stubs (URL preservation gate per always.md full-corpus mandate)

## Phase 0e3: PDF Deep Crawl + Web-Research Enrichment

***CV/RESUME/BROCHURE/REPORT — STRUCTURED FACTS PIPELINE***

Every linked PDF in source (CV, resume, brochure, menu, whitepaper, annual report, fact sheet) MUST be crawled before Phase 1 build, NOT treated as a generic asset (always.md "Every linked PDF" rule).

**Process**: `pdftotext -layout` → LLM-with-strict-JSON-schema extraction → `_pdf_facts.json` keyed by URL.

### JSON schema per CV/resume PDF

```json
{
  "source_url": "...",
  "person": { "name": "...", "title": "...", "email?": "...", "phone?": "...", "address?": "...", "photo_url?": "..." },
  "education": [{ "degree": "...", "institution": "...", "year": "...", "location?": "...", "thesis?": "...", "advisors?": [] }],
  "positions": [{ "title": "...", "organization": "...", "start_year": "...", "end_year?": "...", "location": "...", "description": "...", "achievements?": [] }],
  "publications": "(run through Phase 0e2 resolver)",
  "grants": [{ "title": "...", "funding_org": "...", "amount?": "...", "year": "...", "role": "...", "project": "..." }],
  "awards": [{ "name": "...", "awarder": "...", "year": "...", "citation?": "..." }],
  "teaching": [{ "course": "...", "institution": "...", "years": "...", "level": "..." }],
  "service": [{ "role": "...", "organization": "...", "years": "..." }],
  "professional_memberships": [{ "org": "...", "role?": "...", "years": "..." }],
  "languages?": [],
  "skills?": []
}
```

### Other PDF types

- **Brochure** — `{ services[], pricing?[], contact_block }`
- **Menu** — `{ sections[]: { name, items[]: { name, description, price? } } }`
- **Annual report** — `{ year, executive_summary, financials?, programs[], impact_metrics[] }`

### Web-research enrichment per timeline node (CVs especially)

- For every `position`/`education`/`grant`/`publication`/`award`, run Tavily/Exa search `<institution> <year> <person-name>`
- Enrich with: institution canonical URL, news/press mentions, related papers, grant agency project page, award announcement
- Persist back into the timeline node as `enrichment: { institution_url, related_papers[], news_mentions[], canonical_event_url? }`
- Hyperlink all of it in render

### CV render = interactive Timeline component

On `/about` (or dedicated `/cv` route when CV is a primary credential):

- Vertical chronological line with year markers (left rail)
- Role/event cards on alternating sides
- Concise summary headline + 1-2 sentence body
- Smaller text for supplementary metadata (org name, location, dates) at ≥7:1 contrast
- Every institution/paper/grant/award name hyperlinked to its canonical URL
- Expand-on-click for full details
- Keyboard navigable (Tab through cards, Enter to expand)
- In-viewport `animate__fadeInUp` per card via IntersectionObserver
- ARIA `role="list"` + `aria-label="Career timeline"`
- Reduced-motion: skip stagger, render all cards visible

The lonemountainglobal.com `Vian_CV_Long_4-2-2024.pdf` is the canonical case — Dr. Taryn Vian's 30+ years of WHO/USAID/BU positions, anti-corruption publications, and global health grants become a living timeline, not a buried PDF link.

### Per-prompt context injection

`_pdf_facts.json` is loaded into EVERY downstream content prompt (about-page, team-page, services-page, blog-post-generation, JSON-LD personSchema, FAQ generation) so CV-structured facts back every assertion the rebuild makes about the person/org. Without this, the LLM hallucinates dates/positions and the fact-checker (skill 09 grammar-audit) flags inconsistencies.

**Validator** — `validate-pdf-facts.mjs` asserts every linked PDF in source has corresponding `_pdf_facts.json` entry before container build can start.

## Phase 0e4: Demographic + Page-Set Gap Detection

***REBUILD/OPTIMIZE/ENHANCE — UNION(SOURCE, STANDARD, JEWELS, LOCALES)***

Rebuild prompts MUST emit a superset, never a subset, of source. Three parallel gap-detection passes run after `_research.json` lands, all feeding the build prompt.

See [[source-site-enhancement]] § Union-Output Contract + [[monitor-orchestration]] Known-shortcoming #7 + [[page-set-expansion]] § Org-Type Canonical Floor + [[i18n-by-demographics]] § ACS B16001 Lookup.

### (a) ACS B16001 demographic lookup (locale gap) → `_locales.json`

- **Input**: `_research.json.operations.service_area` (city/county/ZIP set) + `_research.json.operations.geo`
- **Query**: Census API `https://api.census.gov/data/2022/acs/acs5?get=NAME,B16001_001E,B16001_<group>E&for=county:<fips>&in=state:<fips>&key=$CENSUS_API_KEY` for every county touching service area
- **Output**: `{ primary_locale: "en", secondary_locales: [{ code, name, pct_speakers, lep_pct, source_county, apa_citation }], translation_strategy: "full_route_alt|hreflang_only|none" }`
- Per [[i18n-by-demographics]] thresholds (≥5% LEP speakers OR ≥10K absolute LEP population → full translation; ≥2% → hreflang-only stubs)
- Confidence 0.92 (gov_edu source)
- NEVER infer locales from business-name heuristics ("Casa de…" ≠ Spanish locale gate) — Census data is the only source of truth

### (b) Page-set gap diff (route gap) → `_page_set_gap.json`

- **Input**: `_scraped_content.json.sitemap_urls[]` (source URL inventory, post-Squarespace-dedup) + `_research.json.identity.schema_org_type` (org-type → canonical-floor mapping in [[page-set-expansion]] § Org-Type Canonical Floor)
- For each org type (NonProfit, Restaurant, LegalService, MedicalBusiness, Salon, Church, ProfessionalService, SaaS, Portfolio): emit `{ org_type, source_routes: [], canonical_floor_routes: [], missing_routes: [], extra_routes: [], jewel_candidates: [] }` where:
  - `missing_routes` = floor − source (rebuild must ADD these)
  - `extra_routes` = source − floor (rebuild keeps these too, never drops)
- **Example**: NonProfit floor = `/about`, `/programs`, `/impact`, `/team`, `/donate`, `/volunteer`, `/we-need`, `/events`, `/blog`, `/contact`, `/financials`, `/annual-report`, `/faq` — source might only have 6 of these; missing 7 become jewel_candidates

### (c) Jewel content briefs (content gap) → `_jewels.json`

Per [[page-set-expansion]] § Jewel Content Authoring Playbook, each route in `missing_routes` gets a `JewelBrief`:

```ts
{
  route, h1, target_word_count: 1200-2500, keyphrase,
  intent: "info|conversion|trust|local-seo",
  required_sections: [],
  data_dependencies: [refs into _research/_pdf_facts/_corpus],
  image_requirements: { count, type, source_priority },
  jsonld_types: [],
  related_links: [],
  apa_citations_required: [refIds]
}
```

Briefs are dense — each one is the seed prompt for a parallel jewel-content-author agent (see [[monitor-orchestration]] § Parallel-agent playbook).

**Confidence**:

- Jewel content authored from `_pdf_facts.json`+`_research.json` confirmed data = 0.85
- LLM-synthesized with web_search corroboration = 0.70
- Pure inference = 0.50 (flagged in `warnings[]`)

### Persistence

Persist all three to `_research.json.gaps = { locales: ref, page_set: ref, jewels: ref }` so the build prompt loads the union(source, floor, jewels, locales) automatically.

**Validator** — `validate-rebuild-union.mjs` fails build if:

- Shipped sitemap.xml count < `source_routes.length + missing_routes.length`
- OR any `secondary_locales[].translation_strategy = "full_route_alt"` lacks `/<locale>/<route>` siblings

## Phase 0f: Enrichment Sources

- **Google Places** — `GOOGLE_PLACES_API_KEY` → Profile, hours, reviews, photos, rating (confidence 80-95)
- **Yelp Fusion** — `YELP_API_KEY` → Reviews, photos, rating, categories (60-80)
- **Foursquare** — `FOURSQUARE_API_KEY` → Venue photos, tips, hours (65-75)
- **Google CSE** — `GOOGLE_CSE_KEY + CX` → Business-specific web images (40-70)
- **Logo.dev** — `LOGODEV_TOKEN` → Logo image URL (85)
- **Brandfetch** — `BRANDFETCH_API_KEY` → Full brand kit (logo, colors, fonts) (90)

## Output Format

`_research.json`:

```json
{
  "identity": { "name": Conf, "tagline": Conf, "category": Conf, "schema_org_type": Conf },
  "operations": { "phone": Conf, "email": Conf, "address": Conf, "hours": Conf, "geo": Conf },
  "offerings": { "services": Conf[], "menu": Conf[], "pricing": Conf },
  "trust": { "rating": Conf, "review_count": Conf, "reviews": Conf[], "years_in_business": Conf },
  "brand": { "colors": Conf, "fonts": Conf, "personality": Conf, "logo_url": Conf },
  "marketing": { "selling_points": Conf[], "hero_slogans": Conf[], "benefit_bullets": Conf[] },
  "media": { "images": Conf[], "videos": Conf[], "placeholder_strategy": "css_gradient" },
  "seo": { "primary_keyword": Conf, "secondary_keywords": Conf[], "faq": Conf[] },
  "provenance": { "overallConfidence": number, "sectionConfidence": {}, "warnings": [], "version": "v3" }
}
```

## Research for Different Site Types

- **SaaS sites** — Skip Google Places. Research: competitor features (web search), pricing benchmarks, integration ecosystems, trust signals (G2/Capterra ratings), tech stack indicators. Scrape: landing pages, pricing pages, docs, changelog.
- **Portfolio sites** — Minimal API research. Focus on: scraping all project/work pages, extracting case study content, downloading project images, identifying skills/tech mentioned. Client list from testimonials.
- **Non-profit** — Google Places + IRS 990 data (if available). Research: mission statement, impact metrics, volunteer programs, donation platforms, event calendars, partner organizations. Scrape all pages — non-profits often have 20+ pages of valuable content to reorganize.
- **Government/institutional** — Deep scrape required (often 100+ pages). Organize by user intent not org structure. Extract: services offered, contact directories, document libraries, news/press releases, policy documents.
writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall