Breakthrough Content Engine
Give it a goal and it produces a complete content plan. People review only one table; everything else is mechanically derived.
This is a dispatcher. The map below tells you which layer lives in which
file.
Load on demand; do not read everything at once.What makes it different
Most content tools have the shape of “Give me a topic and I’ll give you a draft.” This system reverses that: You give me a business goal, and I first tell you who this run should target, what it should cover, and how many pieces to produce. Only then do we write the copy.
Three things make it different:
- The mix is derived, not copied from a ratio. No empirical research supports any number of content pieces per funnel stage (80/20, 4-1-1, and 70/20/10 are all assertions; 70/20/10, moreover, is a risk allocation misquoted as a funnel allocation). The numbers here therefore come from your capacity, your funnel diagnosis, and your runway, not from a slogan. Capacity sets the ceiling, the goal gives it shape, diagnosis refines it, and format locks it in.
- People review the seed, not the drafts. A seed sheet of 8 to 20 entries is the only human decision point. The content map of roughly 12 to 20 slots is mechanically derived from the seed and can be regenerated. No one should review those twenty-odd slots one by one; scale should serve human judgment, not overwhelm it.
- It says no. If the audience is not large enough, it does not produce a mix; it produces a reach-and-acquisition task list instead. If there is no real material at hand, it does not write a draft; it produces collection instructions. “Not enough material—go get material” is a legitimate, correct output.
When to use it
- ✅ You need to plan an entire content run (one month, a launch, or a stretch of runway)
- ✅ You keep publishing but get no results, and want to know which stage of the funnel is leaking
- ✅ You want to know how many videos and graphic posts this run should include instead of relying on gut feel
- ❌ You only need one piece of copy → do not run the entire pipeline
- ❌ Paid media is the primary channel → the mix logic is completely different; this engine only handles organic content
- ❌ You do not yet have a product, or no one has bought anything → that is a product-validation problem, not a content problem
Four modes
This skill has four ways in. Determine which one applies before proceeding.
| Mode | When | What it does | Load |
|---|
| Plan | “Help me plan this month” or “How many pieces should this run include?” | The main path from objective to finished drafts, ending with a revision feedback loop | modes/plan.md |
| Capture | “I picked up a few things customers said” or “I filled in those NEEDS-CAPTURE items” | Collect real material, assign , and upgrade the evidence tier of a barrier | modes/capture.md |
| Log | When the doorbell names a task, or “Record the results” | Data entry only at 3 / 7 / 15 days, completed in 30 seconds, with no judgment | modes/log.md |
| Review | “Run a review,” or 6 to 8 weeks after a content run ends | Read the data, rerun the diagnosis and mix, and produce proposals rather than conclusions | modes/review.md |
Capture is the engine’s only entry point for real evidence. Everything entering from elsewhere is context.
If you have a second-brain vault
If you can see
99_Meta/structure-doctrine.md
, you have one. If you do, begin by reading
references/_vault.md: scan the Business Wing for brand context, read the
overlay (numbers calibrated by the user take precedence over factory defaults), and understand where things should live.
🛑
The first action in every mode is wayfinding (
§1.5,
run the entire section): use one glob operation to find the
address book and take every path from it; once you have it, continue with the doorbell wiring check in the same section. ⛔ Whether this is the first time or the hundredth time, ⛔ do not guess paths from memory: the names of the Business Wing and brand folders are
chosen by the user, so they cannot be guessed.
⚠️
What comes back from the scan is context, not evidence. The Eight-Pillar Foundation and the brand strategy report may read like evidence—even the quotation marks make them look that way—but testing a locked 100KB strategy produced
0 customer barriers. Therefore, the first action after scanning the vault is a quotation-provenance scan; anything unmarked is treated as constructed.
The engine works without a vault too. Ask the six questions in
as usual. ⛔ A vault is a bonus, not a prerequisite.
Doorbell wiring (during wayfinding, ⛔ not at installation)
If the user has a product-generated command-base skill, the engine will modify the copy the user has already installed (not the product’s template—the template only affects future installs) so that the doorbell recognizes when content is due.
⛔
This does not happen at installation. is a suite manager; it does not execute any instruction inside the skill, and at the moment of installation nobody knows where the user’s home directory is. ⇒
The trigger is the wayfinding step at the start of every mode (
§1.5, “There is one more step after wayfinding”), ⛔
not §8: §8 runs only when the address book does not exist, so users who already have one would never reach it.
⚠️ Product-side doorbell doctrine:
The bell rings only once, and the bell itself names what is due, without throwing a multiple-choice question back to the owner (the product’s
, “One doorbell, two files,” says exactly this). Therefore,
do not add a second bell; simply write
which pieces have reached 3 / 7 / 15 days somewhere the existing doorbell can read it. The bell still rings once and now names one additional item.
⛔ Modify
the copy Claude Code actually loads (
~/.claude/skills/<slug>-command-base/SKILL.md
). In a copy-based installation, the copy in the vault is merely a duplicate; editing it will raise no error but will have no effect. After editing, use grep to read it back and confirm the change. ⛔ Do not write into the doorbell block itself; a product upgrade will overwrite that entire block.
Which branch to use, where to insert it, what to insert, how to rerun without inserting it twice, and how to record a failed connection—all five are documented in
§1.5 and §9. ⛔ Do not act based on memory of this section.
Follow these layers in order while running
| Layer | What it does | Who does it | File |
|---|
| L0 | Collect the objective, deadline, capacity, platforms, and audience size | Ask | |
| L0.5 | Scale gate: if the audience is not large enough, do not produce a mix | Gate | |
| L1 | Funnel diagnosis; produce a stage-pressure profile | Engine suggests | |
| L2 | Mainline recommendations; 2 to 3 cards | Human chooses (decision point 1) | |
| L3 | Mix: how many pieces per stage and format, and which platforms | Engine suggests; human may change the numbers | |
| L4 | Seed sheet | Human reviews (decision point 2, the only review) | |
| L5 | Content map, mechanically derived | Engine | |
| L6 | Select one slot and draft the copy | Engine | → → → |
Load on demand
| What the user wants | What to read |
|---|
| Start a new content plan / “Help me plan this month” | |
| “Why isn’t my content getting views?” / diagnose where the funnel is blocked | |
| Set the mainline / “What angle should I take?” / reframe | |
| “How many pieces should this run include?” / mix / which platforms | |
| Select a content type / “What can I post?” | |
| Produce a seed sheet / review the seed | |
| Derive the content map / how the content pieces connect | |
| Write one piece / how one piece should be structured | |
| How to publish on a specific platform | (choose one by platform) |
| Want a title / opening line / video hook | (read first) |
| The copy reads like AI / pre-publication check | |
| “I have no material to write from” | |
| Run an entire content plan at once / batch processing | |
| Have a vault / override values / where things should live | |
| Change a rule / find out why a rule was decided that way | (authoritative arbitration source; ⛔ to change a rule, edit the source file first, then return to record the ruling and rationale) |
| Numbers and lessons the user has calibrated themselves | The vault’s 04_Methodology/Playbooks/Content-Engine/
|
Rules that apply throughout
These rules appear in many files because they genuinely apply everywhere:
- Brand-name masking test. Cover the brand name. If any competitor could publish the piece, it has failed. It succeeds only if it is immediately recognizable as yours.
- The free-teacher trap. Knowledge is cheap, and AI can generate it on demand; an opinion is valuable. How-to content is especially prone to this trap. Every piece must carry your own method and position.
- Deep ≠ unique. Dig into emotion or identity, and you will often end up with a universal truth that applies to everyone. That is the highest form of homogenization, not insight. Differentiation comes from a reframe (your perspective), not from digging deeper.
- One pitch in Awareness collapses all three stages. If a product appears in awareness-stage content, readers will swipe away immediately. There are no exceptions.
- Only genuine scarcity, never manufactured urgency. At the wrong stage, manufactured urgency is a category error, not a style choice.
- Real memories contain incidental material; fabricated memories do not. If every detail in a draft happens to serve the argument and not one detail is superfluous, it is probably fabricated. That superfluous detail must also come from the same source as the load-bearing details; otherwise, any random irrelevant detail could pass and the rule would be pointless.
In one sentence
Content is slow work. Get real material first, then find an angle, then create the content. What this toolkit can do is ground every step; it cannot retrieve the sentence a customer actually said for you.