Skip to content

An agent organization structured as a company — 15+ departments, 125+ skills, each independently installable, citing the standards and regulators that settle the question. Runs in Claude Code and ChatGPT.

Add Never blocks to marketing and demand-generation

Claudeb973f2057178
diff --git a/plugins/demand-generation/skills/ai-search-optimization/SKILL.md b/plugins/demand-generation/skills/ai-search-optimization/SKILL.md
index ede293f..717a8e5 100644
--- a/plugins/demand-generation/skills/ai-search-optimization/SKILL.md
+++ b/plugins/demand-generation/skills/ai-search-optimization/SKILL.md
@@ -44,3 +44,9 @@ correcting the source, not the assistant.
Click-through will fall on informational queries even as influence rises. Track citation and mention
frequency, and downstream branded search and direct traffic, rather than judging this program on
organic sessions — that metric will say you are losing while you are winning.
+
+## Never
+
+- Optimize for one assistant's current behavior. Retrieval and citation rules change without notice and without a changelog.
+- Assume being crawlable means being citable. Models cite sources that answer a question cleanly, not sources that merely exist.
+- Leave an inaccurate representation uncorrected because it is not on your site. The claim propagates whether or not you own the page it came from.
diff --git a/plugins/demand-generation/skills/app-store-optimization/SKILL.md b/plugins/demand-generation/skills/app-store-optimization/SKILL.md
index 70282dd..516814e 100644
--- a/plugins/demand-generation/skills/app-store-optimization/SKILL.md
+++ b/plugins/demand-generation/skills/app-store-optimization/SKILL.md
@@ -46,3 +46,10 @@ Watch review text for recurring themes — it is the cheapest continuous product
Change one element at a time and let it run a full weekly cycle; app traffic is strongly
day-of-week seasonal. Attributing a lift to the wrong change is worse than not testing.
+
+## Never
+
+- Chase a keyword the app does not deliver on. Installs from a mismatched query become one-star reviews and a worse ranking than you started with.
+- Change metadata, screenshots, and the icon in the same release. Nothing that moves afterward can be attributed.
+- Solicit ratings from a user mid-task. The prompt lands where frustration is highest and the score reflects that.
+- Ignore reviews on the version you just shipped. They are the fastest signal you will get that a release broke something.
diff --git a/plugins/demand-generation/skills/experimentation/SKILL.md b/plugins/demand-generation/skills/experimentation/SKILL.md
index f37eb75..d9828cd 100644
--- a/plugins/demand-generation/skills/experimentation/SKILL.md
+++ b/plugins/demand-generation/skills/experimentation/SKILL.md
@@ -46,3 +46,10 @@ tests a year — spend them on structural questions, not button colors.
Keep a log of every test: hypothesis, result, decision. Without it, teams re-run the same tests every
eighteen months and re-learn the same things.
+
+## Never
+
+- Stop a test because it reached significance early. Peeking until it looks conclusive manufactures the result.
+- Run a test that cannot reach adequate sample size in a reasonable window. Ship the change on judgment instead and say so.
+- Change more than one variable and attribute the outcome to the one you liked.
+- Count a flat result as a failure. A well-run test that rules out a plausible idea has bought information.
diff --git a/plugins/demand-generation/skills/landing-page-cro-expert/SKILL.md b/plugins/demand-generation/skills/landing-page-cro-expert/SKILL.md
index df9211e..ff5d9eb 100644
--- a/plugins/demand-generation/skills/landing-page-cro-expert/SKILL.md
+++ b/plugins/demand-generation/skills/landing-page-cro-expert/SKILL.md
@@ -70,6 +70,12 @@ objections handled, and visual hierarchy supporting the path.
Publish the criteria with the score. A page scored 6/10 with no rubric produces an argument; the
same score with the rubric produces a work list.
+## Never
+
+- Rewrite before diagnosing. Most pages that fail to convert are failing above the fold or off the page entirely, in the traffic.
+- Test a variant you cannot explain the hypothesis for. A win you cannot account for does not generalize.
+- Remove an objection from the page because it is uncomfortable. It does not leave the buyer's head, only your copy.
+
## Return contract
Findings ordered by expected impact, each with the evidence, the specific fix, and effort. The one
diff --git a/plugins/demand-generation/skills/lead-capture/SKILL.md b/plugins/demand-generation/skills/lead-capture/SKILL.md
index 2981964..36b438e 100644
--- a/plugins/demand-generation/skills/lead-capture/SKILL.md
+++ b/plugins/demand-generation/skills/lead-capture/SKILL.md
@@ -52,3 +52,10 @@ Popups work and are widely hated; both facts matter.
Every field costs conversion. Ask for what you need to do the next step and nothing for a later one
— enrichment can fill the rest. Explain any field whose purpose is not obvious, since unexplained
fields are where people abandon.
+
+## Never
+
+- Gate something the buyer can get ungated elsewhere in one click.
+- Ask for a field nobody downstream uses. Every additional field costs conversion and the sales team is not reading it.
+- Trigger a popup before the visitor has seen anything worth staying for.
+- Treat a form fill as intent. Most of them are people who wanted the asset.
diff --git a/plugins/demand-generation/skills/listing-distribution/SKILL.md b/plugins/demand-generation/skills/listing-distribution/SKILL.md
index c667ef4..f027ffd 100644
--- a/plugins/demand-generation/skills/listing-distribution/SKILL.md
+++ b/plugins/demand-generation/skills/listing-distribution/SKILL.md
@@ -52,3 +52,9 @@ set is worse than nothing.
Set a periodic sweep: listings drift out of date, and an aggregator showing your old pricing will
outrank your own page for some queries.
+
+## Never
+
+- List everywhere available. Directories with no traffic cost maintenance and dilute nothing but your attention.
+- Let listing copy drift from the positioning of record. Every stale listing is a competing description of the product.
+- Solicit reviews only after a customer complains, or only from customers you know are happy. Both distort the signal you are buying.
diff --git a/plugins/demand-generation/skills/marketing-analytics/SKILL.md b/plugins/demand-generation/skills/marketing-analytics/SKILL.md
index 4b4fd59..ce896d2 100644
--- a/plugins/demand-generation/skills/marketing-analytics/SKILL.md
+++ b/plugins/demand-generation/skills/marketing-analytics/SKILL.md
@@ -51,3 +51,10 @@ nobody.
Show the metric, its comparison period, and the decision it informs. A number with no comparison is
not information. Where a number moved, the report should say why or say that the cause is unknown —
"unknown" is a legitimate and useful finding.
+
+## Never
+
+- Add tracking before the plan names the events and their properties. Retrofitting a schema onto live data is a migration, not an edit.
+- Report an attribution number without saying which model produced it. The same period looks like different businesses under first and last touch.
+- Change an event definition without versioning it. Every historical comparison silently becomes wrong.
+- Build a dashboard nobody named a decision for.
diff --git a/plugins/demand-generation/skills/seo-strategy/SKILL.md b/plugins/demand-generation/skills/seo-strategy/SKILL.md
index 73f128d..eab9bf9 100644
--- a/plugins/demand-generation/skills/seo-strategy/SKILL.md
+++ b/plugins/demand-generation/skills/seo-strategy/SKILL.md
@@ -83,6 +83,12 @@ Rank fixes by traffic at risk against effort. In practice the order is almost al
problems, then intent mismatch on high-value pages, then internal linking, then structured data,
then everything else.
+## Never
+
+- Chase rankings for a term the page cannot satisfy. The click arrives, bounces, and teaches the engine the page was wrong.
+- Restructure a site without a redirect map. Lost link equity does not come back on its own.
+- Judge a change in the first fortnight. Search feedback is slow enough that early movement is usually noise.
+
## Return contract
Findings by severity with the evidence for each, the sequenced fix list, expected impact and when it
diff --git a/plugins/marketing/skills/behavioral-marketing/SKILL.md b/plugins/marketing/skills/behavioral-marketing/SKILL.md
index 8b22bc0..185e4d1 100644
--- a/plugins/marketing/skills/behavioral-marketing/SKILL.md
+++ b/plugins/marketing/skills/behavioral-marketing/SKILL.md
@@ -69,3 +69,9 @@ buy anyway.
- **Effects are contextual and interact.** Published effect sizes are directional, not predictions.
Treat every application as a hypothesis to test rather than a known quantity — this is where
behavioral marketing most often overreaches.
+
+## Never
+
+- Deploy an effect without being able to name the mechanism. A tactic borrowed without its mechanism fails silently in a context that differs in ways nobody checked.
+- Manufacture scarcity, urgency, or social proof that is not true. It works once and costs the relationship permanently.
+- Treat a published effect size as a guaranteed lift. Those numbers come from populations and contexts that are not yours.
diff --git a/plugins/marketing/skills/brand-voice/SKILL.md b/plugins/marketing/skills/brand-voice/SKILL.md
index 3976cb8..fe2f163 100644
--- a/plugins/marketing/skills/brand-voice/SKILL.md
+++ b/plugins/marketing/skills/brand-voice/SKILL.md
@@ -46,3 +46,9 @@ Include a **do-not** list. Voice guides fail on what they permit, not what they
Draft two paragraphs and read them to the person cold. If they say "close, but I wouldn't say it
that way," ask exactly what they would say — that correction is the most valuable data in the
process.
+
+## Never
+
+- Build a voice guide from copy the person did not write. Agency and ghostwritten material describes someone else's voice.
+- Describe voice in adjectives alone. "Friendly but authoritative" is unusable without paired examples of what passes and what fails.
+- Apply a voice guide to a format it was never sampled from. How someone writes an email is not how they write a keynote.
diff --git a/plugins/marketing/skills/chief-content-officer/SKILL.md b/plugins/marketing/skills/chief-content-officer/SKILL.md
index fb28af1..aa3738b 100644
--- a/plugins/marketing/skills/chief-content-officer/SKILL.md
+++ b/plugins/marketing/skills/chief-content-officer/SKILL.md
@@ -80,3 +80,10 @@ piles: **update** (performing, stale), **consolidate** (several thin pieces on o
(no traffic, no links, no strategic value).
Most libraries have more value in the promote pile than in anything unwritten.
+
+## Never
+
+- Start production before the brief names the argument. An idea is a title; a piece needs a claim.
+- Fill the calendar to capacity. A pipeline with no slack cannot absorb the one piece worth dropping everything for.
+- Choose the format before the argument. That is how a topic that wanted six hundred words becomes a video series.
+- Retire a piece that has inbound links without redirecting it.
diff --git a/plugins/marketing/skills/content-strategy/SKILL.md b/plugins/marketing/skills/content-strategy/SKILL.md
index c6faa96..73d1c9b 100644
--- a/plugins/marketing/skills/content-strategy/SKILL.md
+++ b/plugins/marketing/skills/content-strategy/SKILL.md
@@ -48,6 +48,12 @@ everything on traffic is why content programs drift toward the reach end and sta
Set a review point where a topic that is not working gets dropped. Content strategies fail by
accumulation.
+## Never
+
+- Claim a territory you cannot sustain publishing on. Withdrawing from one costs more than never entering it.
+- Commit to a platform without deciding who runs it every week.
+- Judge a content program on traffic when the objective was demand.
+
## Return contract
Territory, audience, the four-way format budget, cadence, platforms with rationale, first ten pieces
diff --git a/plugins/marketing/skills/marketing-campaign-planner/SKILL.md b/plugins/marketing/skills/marketing-campaign-planner/SKILL.md
index 61f1172..baf8019 100644
--- a/plugins/marketing/skills/marketing-campaign-planner/SKILL.md
+++ b/plugins/marketing/skills/marketing-campaign-planner/SKILL.md
@@ -80,6 +80,13 @@ contact with a considered buyer.
- What happens if it works far better than expected — can delivery, support, and inventory absorb
it?
+## Never
+
+- Run a campaign carrying more than one story. A second message halves the first.
+- Add a channel because it is available. Every channel added is one somebody has to run properly.
+- Set a launch date before the assets that gate it have named owners.
+- Commit to an objective you cannot measure with a number you already collect.
+
## Return contract
The story in one sentence, objective and leading indicator, channels with sequencing rationale,
diff --git a/plugins/marketing/skills/marketing-copywriting/SKILL.md b/plugins/marketing/skills/marketing-copywriting/SKILL.md
index 2ab540b..e8a2744 100644
--- a/plugins/marketing/skills/marketing-copywriting/SKILL.md
+++ b/plugins/marketing/skills/marketing-copywriting/SKILL.md
@@ -49,3 +49,9 @@ Edit in passes; doing them at once does none of them.
Empty openers ("In today's fast-paced world"), we-focused framing where you-focused works, hedges
that weaken a true claim, feature lists with no consequence attached, and any adjective a competitor
could equally use.
+
+## Never
+
+- Write before you know what the reader should do next.
+- Edit for elegance before the argument is right. Polished copy for the wrong claim is the more expensive mistake.
+- Approve copy nobody read aloud. Rhythm problems are inaudible on the page.
diff --git a/plugins/marketing/skills/newsletter-writer/SKILL.md b/plugins/marketing/skills/newsletter-writer/SKILL.md
index 4ed18fb..837e02a 100644
--- a/plugins/marketing/skills/newsletter-writer/SKILL.md
+++ b/plugins/marketing/skills/newsletter-writer/SKILL.md
@@ -81,3 +81,9 @@ read-aloud pass catches what silent reading never does.
One issue contains several short-form posts, a talk track, and often a longer piece. Plan which
before writing, so the sections that should stand alone are written to.
+
+## Never
+
+- Write a subject line the opening does not pay off. It buys one open and costs the next.
+- Sell in an issue that gave nothing else. The trust spent is not recovered by the next good one.
+- Send an offer to the whole list when it applies to part of it.
diff --git a/plugins/marketing/skills/public-relations/SKILL.md b/plugins/marketing/skills/public-relations/SKILL.md
index b352111..a2c4f30 100644
--- a/plugins/marketing/skills/public-relations/SKILL.md
+++ b/plugins/marketing/skills/public-relations/SKILL.md
@@ -46,3 +46,10 @@ explicitly agreed in advance — and even then, assume it may not hold.
guarantee a second story.
- Correct genuine errors privately, with evidence, and specifically. Vague complaints about tone
achieve nothing.
+
+## Never
+
+- Pitch a story that is only news to you. A funding round, a hire, and a redesign are not news outside the company.
+- Send a pitch to a journalist whose recent work you have not read.
+- Enter an interview without deciding in advance what you will not say.
+- Respond to a damaging story before establishing whether it is accurate.
diff --git a/plugins/marketing/skills/social-post-craft/SKILL.md b/plugins/marketing/skills/social-post-craft/SKILL.md
index 9aec8e6..be5e95e 100644
--- a/plugins/marketing/skills/social-post-craft/SKILL.md
+++ b/plugins/marketing/skills/social-post-craft/SKILL.md
@@ -51,3 +51,9 @@ most platforms strip formatting.
- Would you send this to one person you respect? If not, do not send it to everyone.
A draft that fails these gets rewritten, not published with a caveat.
+
+## Never
+
+- Publish a post whose first line only works after the reader expands it.
+- Reuse one format across platforms unchanged. Line breaks and truncation differ, and the post breaks in ways the composer does not show you.
+- Post a link without deciding whether the platform suppresses them.
diff --git a/plugins/marketing/skills/video-content/SKILL.md b/plugins/marketing/skills/video-content/SKILL.md
index 46537f6..8ecbb3e 100644
--- a/plugins/marketing/skills/video-content/SKILL.md
+++ b/plugins/marketing/skills/video-content/SKILL.md
@@ -49,3 +49,9 @@ uniformly intense video is as tiring as a flat one.
Plan derivatives before filming, not after. Segments intended to stand alone get shot to stand
alone. Extracting short-form from a video not built for it produces clips that need context they do
not have.
+
+## Never
+
+- Produce before the title and thumbnail exist. Packaging decides whether anything you shot gets seen.
+- Open with a preamble. The first seconds are where an audience is lost, not where it is warmed up.
+- Diagnose a video on view count alone. Impressions, click-through, and retention fail differently and need different fixes.
diff --git a/plugins/marketing/skills/visual-content/SKILL.md b/plugins/marketing/skills/visual-content/SKILL.md
index 808459e..5562dae 100644
--- a/plugins/marketing/skills/visual-content/SKILL.md
+++ b/plugins/marketing/skills/visual-content/SKILL.md
@@ -50,3 +50,9 @@ Write the prompt as a brief: subject, composition, lighting, palette, mood, and
constraints. Generate one concept per image. Then check the output actually renders any embedded
text correctly and matches the brand palette — generated imagery drifts, and drifted brand color is
worse than no brand color.
+
+## Never
+
+- Choose a visual format before knowing whether the content is a sequence, a comparison, or a single claim.
+- Design at full size for something that will be seen at thumbnail size.
+- Put load-bearing information in an image alone. It is unsearchable, untranslatable, and invisible to a screen reader.
diff --git a/plugins/marketing/skills/youtube-producer/SKILL.md b/plugins/marketing/skills/youtube-producer/SKILL.md
index 7b44c7d..240f713 100644
--- a/plugins/marketing/skills/youtube-producer/SKILL.md
+++ b/plugins/marketing/skills/youtube-producer/SKILL.md
@@ -87,3 +87,10 @@ different kinds of value teaches the audience nothing about whether to click the
Plan repurposing before filming — segments meant to stand alone as short-form get shot to stand
alone.
+
+## Never
+
+- Produce an idea whose title you cannot write. If the title is hard, the idea is not yet clear.
+- Judge a video on its first day. The signal that matters accrues over weeks.
+- Change what the channel is about because one video underperformed.
+- Fix retention by cutting faster when the problem is that the promise was wrong.