A new session shouldn’t pitch the same ideas we rejected last week. We’ll let Claude Code read the campaign’s decision history before Kabo brings in fresh public references. Our example is a sewing educator choosing the next beginner lesson; the same approach fits any channel with an archive and a clear audience progression.
Kabo, an AI agent for creator research, adds fresh references. Claude Code reads the learner level, covered lessons and rejection reasons before proposing another topic.
Prepare the title history and rejected ideas
[ ] audience.md — Learner level and the next skill we want to unlock. [ ] covered.csv — Published titles and what each lesson already teaches. [ ] rejected.md — Earlier ideas with reasons, including temporary constraints. [ ] assets.csv — Samples, tools and demonstration material we already own. [ ] brief.md — This session’s filming time, topic boundaries and output request.
2. Connect Kabo, then check the project context
Open the content project, install Kabo for Claude Code and complete sign-in. For a direct server connection, follow the setup guide.
3. Read the rejection log before the brainstorm
A sewing educator manages beginner lessons and keeps a backlog of rejected ideas. Their audience can use a basic machine, but is still learning how to recognize construction mistakes.
Have Claude Code summarize the learner’s current skills and the reason earlier ideas were rejected. Keep that summary in the brief, not as a permanent rule against ever covering advanced work. A campaign-level constraint should be easy to update when the audience or available assets change.
Read audience.md, covered.csv, rejected.md and assets.csv. Summarize what our audience already knows, what we covered and why earlier ideas were deferred. Distinguish lasting rules from this session’s constraints before researching.
4. Research what the public examples make understandable
Use Kabo to find relevant recent sewing videos and enrich selected examples where account context is useful. Ask which learner problem a reference addresses and what evidence makes the explanation clear. A dramatic transformation is not necessarily a suitable next lesson for someone who has only sewn straight seams.
5. Classify candidates as test, revise or defer
Ask Claude Code to write ideas.csv with learner level, original contribution, owned sample, source notes and decision reason. Have it compare each title against covered.csv. Review ambiguous candidates before expanding them into briefs. Keep the scores as your rubric-based editorial judgments, not view predictions.
A rejection log needs a reason. “No jackets” becomes an accidental forever rule. “No fitted jacket this month: beginner audience and no fitting model” leaves a useful path back later.
Use the Claude Code comment workflow when the learner obstacle needs a closer look before choosing the next lesson.
Use Kabo to research recent public [topic] examples in [date window]. Propose 5 distinct teaching ideas classified as test, revise or defer. Include learner level, viewer question, our contribution, owned sample, missing proof, source links and decision reason. Check every idea against covered.csv. After I choose, write one brief; don’t turn temporary limits into permanent CLAUDE.md rules.
An illustrative three-way editorial decision
For the sewing channel, the shortlist should show different decisions, rather than five equally enthusiastic pitches:
- Test: Why Your Tote Bag Twists Before You Sew the Last Seam
- Proof: Compare the owned panel samples and mark alignment.
- Revise: Three Zipper Mistakes → limit to one visible beginner problem.
- Defer: Tailored Jacket Fit → exceeds the agreed learner level.
- Evidence: Attach relevant returned source links to each candidate.
6. Write only the selected teaching brief
Choose the twisting-tote angle and request a short script outline with the mistake, the visible diagnosis and the correction you can demonstrate. Update the rejection log only if you made a durable decision. This keeps a single planning session from turning every temporary constraint into an unexplained permanent instruction.
The zipper idea introduces three techniques at once. Keep one visible beginner mistake and remove the advanced finish. Update the brief, but do not permanently ban zipper topics.
Approve one lesson at the intended learner level
Open the backlog in Google Sheets for an editorial review. Take the selected mistake sample to filming and preserve the lesson-level note with the script so the finished video does not drift into an advanced tutorial.
A missing source belongs in the research dossier. An accepted lesson belongs in the script project, with its learner-level constraint attached.
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.