# Brand Form > Fill out a brand's creator/collab application form (Google Forms, Typeform, etc.) using real channel data. Use when Justin pastes a form link from a brand that replied to a sample request — "help me fill this out", "brand sent me a form", "collab form", "creator application". Scrapes the real questions, pulls verified YouTube numbers, and drafts every answer with nothing invented. - Skill: `thejdubb02/brand-form` (Agent Skill) - Install (CLI): `npx skillmds@latest add thejdubb02/brand-form` - Raw SKILL.md: https://api.skillmd.com/api/skills/thejdubb02/brand-form/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: thejdubb02 (https://skillmd.com/u/thejdubb02) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/thejdubb02/brand-form --- # Brand collab form filler A brand replied to a sample request and sent an application form. Turn the form link into a copy-paste answer sheet built on **verified** numbers. ## The one rule that matters **Every factual answer must trace to a real source.** Subscriber counts, average views, "do you make X content" — these are screening questions, and a brand that ships a sample on an inflated number and gets a 4-view video is a brand that never works with us again. The reach is what it is. The job is to find the *true* strongest angle, not a flattering one. If the honest answer is weak, say so to Justin and let him decide — never quietly round it up. Never invent: shipping address, phone number, or any number you could not pull. Leave those as `[NEED FROM JUSTIN]` and list them at the top of the answer sheet. ## STEP 1 — scrape the real questions Do not eyeball the form in a browser. Google Forms embeds every question in a JS blob: ```bash curl -sL "" -o /tmp/form.html grep -o 'FB_PUBLIC_LOAD_DATA_ = .*' /tmp/form.html | head -c 8000 ``` Read the structure: each question is `[id, "Question text", null, TYPE, [[...options...]]]`. TYPE: `0`=short text, `1`=paragraph, `2`=multiple choice (pick one), `4`=checkboxes (pick many). For multiple choice / checkboxes the exact option strings are in the nested array — **quote them verbatim** in the answer sheet so Justin just clicks, no interpretation. Not a Google Form (Typeform/Jotform/Airtable)? Fall back to WebFetch, then the browser MCP (`mcp__winchrome__navigate` + `snapshot`) if it's fully JS-rendered. ## STEP 2 — pull the real numbers Channel IDs live in `/opt/relay/youtube.py` (`commquest`, `jdubb`, `dubb_outdoors`, `everyday`). ``` mcp__youtube-analytics__channel_analytics(channel="cq", start_date=..., end_date=...) ``` Gives views, watch time, subs gained/lost, traffic sources for a window. **Analytics API — separate quota, usually works.** `list_videos` / `top_videos` hit the **Data API quota, which is routinely exhausted** (see the Shorts-via-OmniSocials memory). When they 403, scrape the public channel page instead: ```bash curl -sL -A "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 Chrome/120.0 Safari/537.36" \ "https://www.youtube.com/channel/" > /tmp/ch.html grep -oE '.{90}subscribers' /tmp/ch.html | head -6 # context disambiguates OUR channel from sidebar channels grep -oE '\{"content":"[0-9.,KM]+ (views|videos)"' /tmp/ch.html | head -6 ``` ⚠️ The bare `grep -oE '[0-9.,KM]+ subscribers'` returns **recommended sidebar channels too**. Always grep with leading context and match the handle (`@Comm_Quest`) before believing a number. **Average views per video is the trap.** Channel-total views ÷ 90 days looks great and means nothing when the catalog is thousands of evergreen videos living on search. Get *recent per-upload* views from the channel page grid (`{"content":"N views"}`) — that is what the brand is actually asking about, and it is usually far lower than the channel total implies. ## STEP 3 — pick the right channel (ask Justin) One channel, one topic — per `ventures/CLAUDE.md`, we do **not** cross-post. The channel with the biggest audience is often not the channel the product belongs on. When those two differ, that is a real decision and it is **Justin's**, not yours: present both with their true numbers and a recommendation, then wait. ## STEP 4 — write the answer sheet Write to the scratchpad as markdown, in **form order**, each question with: - the question verbatim - the answer, ready to paste (for choice questions, the exact option text to click) - a one-line `why:` for anything a human might second-guess Head the file with `[NEED FROM JUSTIN]` items. End with the honest-read section: what this application's real weaknesses are and what genuinely strongest angle the free-text fields lean on. Free-text fields ("anything else you'd like us to know?") are where a modest channel wins or loses. Lead with a true, specific, verifiable asset — evergreen search traffic, a storefront, turnaround speed — never adjectives.