AGENT SKILL

product-business-kit

Use when making a product a full business package.

License 许可
MIT
最近更新
2026-08-16
Stars
★ 0
Forks
0

01 前言 Front Matter

name
product-business-kit
author
Hermes curator
license
MIT
version
1.0.0
metadata
{ "hermes": { "tags": [ "business", "market-research", "brand", "pitch-deck", "content", "audience", "go-to-market" ], "related_skills": [ "brand-symbol-design", "pitch-deck-builder", "underoos-briefing" ] } }
platforms
[ "linux", "macos", "windows" ]
description
Use when making a product a full business package.

02 资源文件 3

03 SKILL.md 正文

Product Business Kit (full package)

When to Use

  • "Make X a full brand and business", "prepare the pitch deck and all the assets", "find market and the pain point and the customers", "find me audience for my product".
  • Ibrahim's products (Siafu, Olive, JengaConnect, Ghost, KukuPlus...) that need the complete package: research → brand → deck → content → audience → delivery index.

Build a complete business package for a product from repo truth: research → brand → deck → content → audience → delivery index.

Workflow (8 phases, in order)

  1. Product truth first — read the repo's README + design docs + BOM before any research. Never invent specs; quote the design doc (dimensions, BOM cost, sensors, algorithm names, current status table). This grounds every later claim.
  2. Parallel market research — batch 6–8 firecrawl searches in ONE turn: market size (category + adjacent), competitor pricing (name + $ per unit + business model), regional problem statistics, regional funding, education/adjacent markets. Distill to a table, don't dump raw results. Always note source + year per figure.
  3. Write BUSINESS-RESEARCH.md — fixed structure: exec summary table → product truth (specs, status) → market size + problem quantified → competition (price-gap table: competitor price vs our BOM) → 5 customer segments each with pain points + how to sell → revenue model (3 phases) + unit economics → go-to-market (90-day plan, risks) → funding map → KPI targets. Flag figures that vary across firms.
  4. Brand kit — delegate to brand-symbol-design skill: 3 directions → geometry-first SVG → color/mono/reversed variants → size-test strip (96→24px) → pixel-probe verification (all probes must PASS before delivery). Brand doc: name story (incl. local proverb if the name is Kiswahili), taglines, voice rules, palette, symbol rationale + usage rules.
  5. Pitch deck — delegate to pitch-deck-builder: 12-slide arc (Title → Problem → Market → Why-now → Solution → Product → Revenue → Competition → Why-now-rails → Customers → Roadmap → Close), dark title/close, footer citations on every content slide. Run qa_pptx.py; then re-export a PDF preview explicitly (see pitfall below).
  6. Content pack — 4 content pillars (problem / technology / story / business), 15 ready-to-post pieces spanning X threads, TikTok/Reel scripts, LinkedIn posts; platform cadence table; 30-day calendar; caption skeleton; brand-voice rules.
  7. Audience map — segments → channels matrix (where each segment lives online + how to reach), communities to join, influencer/account engagement targets, events calendar, copy-paste outreach email templates per segment, funnel + KPIs, and a "where the money actually is" priority order.
  8. README index + delivery — README.md table of every file, 3-line pitch, next-actions checklist (with real deadlines). Deliver with absolute paths + MEDIA: links for the key files (research doc, deck pptx+pdf, symbol png).
  9. Landing page (promo sites) — single-file 3D scroll page via scroll-driven-3d-websites (hero-variant pattern: OrbitControls + procedural product model + entrance cascade + gsap.matchMedia() for mobile/reduced-motion). Use the product's Phalanx-style symbol as inline-SVG favicon/nav/logo data URIs. If the user supplies a Sketchfab embed, include it with attribution intact. Verify: node new Function syntax check on the module script, serve on localhost, open_preview + read_preview to confirm render. Ship as landing/index.html in the package.
  10. Google Drive data-room sync — search for the existing data room FIRST (drive search "data room"), respect its numbered-folder structure, add the product as one self-contained project folder. Generate scripts/sync_drive.sh (pattern in references/drive-dataroom-sync.md, starter in templates/sync_to_drive.sh), run it, then VERIFY by listing each folder back from Drive. Add the folder link + resync command to the README and the folder IDs to memory.

Pitfalls

  • qa_pptx.py deletes its PDF — it renders to a temp dir (mkdtemp) that's removed on exit, so the deck folder gets no PDF. To deliver a PDF preview: re-run the LibreOffice conversion explicitly with --outdir <deckdir> (soffice wrapper from the pptx skill: env -u PYTHONPATH python3 ~/.hermes/skills/pptx/scripts/office/soffice.py --headless --convert-to pdf --outdir . Deck.pptx), then pdftotext grep for key claims to verify.
  • Vision provider may reject images (unknown variant 'image_url') — do NOT retry in a loop. Fall back to deterministic pixel probes (brand-symbol-design has the script pattern) + text-layer checks.
  • Dates rot — funding deadlines (e.g. Milken-Motsepe Aug 13 2026) and market figures age. Always carry source + date with every number; re-verify deadlines before quoting in a fresh session.
  • Check the vault first — search_files the Maarifu vault for an existing product note/brief before starting; the pitch-deck-builder skill mines 01. Companies/<Product>/ and 08 Research Lab/Market Research/.
  • Keep the generators — build_symbol.py and build_deck.js stay in the package dir so revisions regenerate from the same script (brand-symbol-design + pitch-deck-builder both require this).
  • Memory is the pointer, not the package — record only the package path in memory (~/Projects/<product>-business/); the docs themselves live in the package.
  • Drive JSON is a dict, not an object — parsing google_api.py output in bash helpers: json.load(sys.stdin).id raises AttributeError: 'dict' object has no attribute 'id'; use subscript json.load(sys.stdin)['id']. Write the helper right the first time.
  • Drive rate limits on rapid loops — tight sequential drive search/upload calls return EMPTY responses → JSONDecodeError: Expecting value. Space calls (sleep 2–3 between) or batch per command.
  • Verify by listing back — "upload ok" is not proof. After sync, query each folder ('<folder_id>' in parents raw query) and confirm counts/names from Drive itself before telling the user it's synced.
  • Trash, never permanent, for test folders — if debugging left a stray folder, drive delete (default trash) is reversible; don't use --permanent on an accident.

Support files

  • references/siafu-package-v1.md — Siafu v1 case: verified market figures with sources, competitor pricing, problem stats, segment structure, funding map, deliverable paths.
  • references/drive-dataroom-sync.md — the Drive data-room sync workflow: folder discovery, structure rules, sync-script pattern, verification, rate-limit + JSON-parse gotchas.
  • templates/sync_to_drive.sh — known-good idempotent sync script (folder creation + upload) to copy and adapt per product.

04 同仓库更多 Skills