A strong opening still needs to sound like us and promise something we can deliver. We’ll use Claude Code as a draft reviewer: flag a specific overclaim, keep the useful shot direction and record why the approved version changed.
Kabo, an AI agent for creator research, can clarify a video-specific question. Claude Code checks the opening against our voice and claim rules, then records a minimal revision.
Decide whether the promise belongs in our voice
Our balcony-gardening channel has a practical voice: show what we do, avoid guaranteed growing results. A punchy first draft says “Never lose another seedling.” The footage only shows tray labels and a shelf.
The problem here is not finding three louder ways to begin. We need to identify the unsupported promise, preserve the useful image and revise the few lines it affects.
Turn the voice brief into a small review checklist
Claude Code can read voice.md and apply it to a draft. The rules need observable tests. “Sound authentic” gives us little to review; “do not promise a result our demonstration does not measure” tells us what to flag.
Add the agreed checks to voice.md alongside approved writing examples, so the audit has both rules and a real voice sample.
State the concrete problem our footage demonstrates. Avoid guaranteed germination, yield or future outcomes. Prefer a visible action to a broad superlative. Keep the maker’s ordinary speaking style. Flag unsupported claims before rewriting. When revising, identify which rule each change addresses.
Prepare the materials for this Claude Code task
Preserve the original draft so a reviewer can see what changed.
[ ] original-hook.md — Exact narration and shot notes. [ ] voice.md — Approved examples and the short review rules. [ ] claim-support.md — What our observations establish, and what they do not. [ ] shots.md — Available tray, label and shelf footage. [ ] hook-review/ — Audit notes, revised opening and a version note.
Connect Kabo
Open the content project, install Kabo for Claude Code and complete sign-in. For a direct server connection, follow the setup guide.
Ask for findings before a new opening
The first deliverable is a compact audit: passage, relevant rule, support and required action. For our example, “Never lose another seedling” fails because the supplied material demonstrates organization, not future plant survival. The close-up label instruction can stay.
If the supplied script and support notes answer the question, we do not need a video analysis. Propose one only for a specific unanswered public-video issue; a new saved Kabo analysis requires Pro and confirmation.
Read original-hook.md, voice.md, claim-support.md and shots.md. Return a short review with exact passage, relevant rule, support available and recommended action: keep, clarify or remove. Separate a factual overclaim from a stylistic preference. Do not rewrite the whole opening or start a new Kabo analysis.
Review a minimal before-and-after
An illustrative revision:
- Before: “Never lose another seedling. Here is the perfect seed-tray setup.”
- After: “By next week, these trays will look alike. Here is how I keep track of what I planted.”
- Kept: The label close-up and pullback to the small shelf.
- Reason: Replace an unmeasured outcome with the organization problem we can show.
This gives the maker something specific to approve. We have not promised that the revised opening will perform better.
Apply the review decisions I approved. Keep the label close-up and the transition into the planting sequence. Replace unsupported guarantees with the actual organization problem. Save hook-review/hook-v2.md, preserve original-hook.md, and show before/after lines with the rule and reason for each substantive change.
Try the line in our own voice
Read the proposed opening over the label close-up. Does “By next week, these trays will look alike” sound like something we would actually say? Can the viewer read the label before we move on? A safe sentence can still be awkward to deliver.
Record a quick phone rehearsal. Give Claude Code the exact phrase we stumbled over and our preferred natural wording. Keep the demonstrated organization problem; do not expand this into a search-and-replace of the whole project.
In the opening we approved, I stumbled on [exact phrase]. My natural wording is [how I would say it]. Rewrite only those spoken lines, preserving the organization problem and label close-up. Offer one read-aloud version and explain any meaning changed. Do not audit the full script or rewrite the caption.
Approve the spoken opening beside the shot
Keep the version the maker can say comfortably and the label the viewer can read. Save the accepted lines beside original-hook.md, then let the editor make the cut.
If the review reveals a factual change that also affects the title, demo or recap, continue with the Claude Code script revision guide. That project-wide correction is a separate task from this opening rehearsal.
Worked examples are illustrative. Product screenshots show existing Kabo web results, not a run of the example in this guide. Use your own materials and the evidence returned in your task.