yt-launch

name	yt-launch
description	Use when the user wants to set up monitoring for a freshly published YouTube video on the Enterprise Vibe Code channel. Tags the video into a cohort, writes a launch hypothesis to the insights ledger, and installs the daily launch-watch LaunchAgent. Triggers on "launch", "track this video", "monitor this drop", "watch the new one", "kick off launch monitoring".
user_invocable	true

yt-launch

Set up a launch-window monitoring stack for a single video. Generalizes the manual sequence used for the Manifesto (H9HR4_Qz-Z0) into a one-shot command.

When this fires

User just published (or is about to publish) a video and wants to track its launch. Triggers include:

• /yt-launch <url> (explicit)
• "Set up monitoring for this video"
• "Track this drop / launch / new one"
• "Watch this for the next few days"

If the user wants to A/B test the title or thumbnail, run /yt-ab separately AFTER this. They're orthogonal.

Args

$1 — YouTube video URL OR bare 11-char ID (required).

Steps

1. Resolve the video ID

Accept any of:

• H9HR4_Qz-Z0 (bare ID)
• https://www.youtube.com/watch?v=H9HR4_Qz-Z0 (full URL)
• https://youtu.be/H9HR4_Qz-Z0 (short URL)
• URLs with extra params (&t=42s, &list=...)

2. Refresh data so the video is captured

cd /Users/mikelady/dev/youtube_analytics
go run . fetch          # 1-2 min; idempotent if already current

This pulls the new video into data/videos.json. If the user just published, the video may not be there yet — fetch should pick it up.

3. Show the baseline

go run . video <id>

Read the deep-dive output. Capture:

• Title
• Published timestamp
• Initial views/likes/duration
• Whether --all analytics dimensions are populated yet (likely sparse for a brand-new video)

4. Ask for cohort + hypothesis

Use AskUserQuestion to confirm two things — but propose smart defaults so the user can usually accept:

Cohort: default evergreen-explainers for substantive long-forms. Offer the existing list:

go run . cohort list

The user can pick an existing one or say "new" and you'll need to add it to data/cohorts.yaml first.

Hypothesis prediction: default to:

<title> launch — predict ≥500 views, ≥30% retention, ≥5 subs/1K UNSUBSCRIBED, ≥10% RELATED_VIDEO traffic by <today + 7 days>.

Adjust the numbers based on channel-typical performance you can see from cohort report for the chosen cohort.

5. Idempotency check

Before making changes, check whether the video is ALREADY set up:

• Is it in data/cohort_assignments.json for the chosen cohort? jq '.\"<id>\"' data/cohort_assignments.json
• Is there an active hypothesis citing it? grep -l "<id>" data/insights/*.md
• Is the launch-watch LaunchAgent already pointing at it? launchctl list | grep yt-launch-watch and check the plist

If all three are present and current, report "already monitored" and stop. Don't double-assign.

6. Execute the setup

# Tag into cohort
go run . cohort assign <cohort> <id>

# Append hypothesis to this Monday's insights file
THIS_MONDAY=$(date -v-mon +%Y-%m-%d 2>/dev/null || date -d 'last monday' +%Y-%m-%d)
INSIGHTS_FILE="data/insights/$THIS_MONDAY.md"
[ -f "$INSIGHTS_FILE" ] || go run . insights new "$THIS_MONDAY"

Then EDIT the insights file directly (using the Edit tool) to append a new hypothesis to the frontmatter hypotheses: list. Use this exact YAML shape:

- id: h-<this-monday>-launch-<id-prefix>     # e.g. h-2026-04-27-launch-H9HR4
  cohort: <cohort>
  prediction: "<the prediction text>"
  evidence_video_ids:
    - <id>
  metric: launch_week_composite
  direction: up
  evaluate_after: "<today + 7 days>"

And append a markdown section below the frontmatter explaining the hypothesis (cite baseline numbers from step 3).

7. Install the launch-watch LaunchAgent

make launch-watch-install VIDEO=<id>

This is the daily 9 AM + 9 PM monitor. It REPLACES any prior launch-watch agent — only one video at a time. If a prior video was being watched, mention this to the user before clobbering.

8. Commit + push

git add data/cohort_assignments.json data/insights/<this-monday>.md
git commit -m "Set up launch monitoring for <id> (<title>)"
git push

The pre-commit hook will run go vet + go test. (No Go files touched, so this is fast.)

9. Report to user

Confirm what's running:

• ✓ Cohort: tagged into <cohort> (now N members)
• ✓ Insights: hypothesis h-<date>-launch-<prefix> written, evaluates <eval-date>
• ✓ LaunchAgent: com.mikelady.yt-launch-watch watching <id>, fires 9 AM + 9 PM local
• Logs: ~/Library/Logs/yt-launch-watch/<id>/
• Mention: run /yt-ab <id> to set up native title testing in YouTube Studio
• Mention: make launch-watch-uninstall after the launch window stabilizes (~5 days)

What this skill does NOT do

• Doesn't drive YouTube Studio. Title/thumbnail testing is /yt-ab's job.
• Doesn't fetch fresh analytics. Brand-new videos have sparse data anyway; the daily LaunchAgent handles ongoing capture.
• Doesn't grade hypotheses. The Sunday weekly review handles grading once evaluate_after passes.
• Doesn't auto-detect launches. The user invokes this manually when they ship.

Reference

• Cohort definitions: data/cohorts.yaml
• Insights ledger format: data/insights/<date>.md (YAML frontmatter, see existing entries)
• Primitive Test: docs/primitive-test.md — this skill is pure orchestration of existing CLI primitives; cognition (cohort choice, hypothesis wording) is the model's job.
• Channel: Enterprise Vibe Code (channel ID UCyxYcJWs5WstpM2h8KjabUw).

This skill lives in the repo

Symlinked from ~/.claude/skills/yt-launch → /Users/mikelady/dev/youtube_analytics/skill/yt-launch. Edit in the repo; symlink picks up changes.