Content Strategy

Content Strategy Builder & Auditor

An interactive skill that either builds a documented content strategy from scratch or audits an existing content program, using the same seven-part framework in both directions. The end product is a markdown file: a strategy with a first quarterly content hub, or a scorecard with ranked gaps and fixes.

The framework rests on one observation about B2B sites: people who land on a blog post are looking for information, not a vendor, so they almost never turn into leads directly. What blog content does do is earn links. Links raise the site's authority. Authority is what lets the service pages rank for the phrases buyers actually search. And service-page visitors are the ones who convert. So the strategy starts at the sales page and works outward, and every content piece has a job: earn links, build relationships with people who publish, or answer a question that's stalling a deal.

Framework: Andy Crestodina, Orbit Media, The B2B Content Strategy Framework. Everything in this skill is a distillation in our own words. Statistics keep their source links in the reference files.

Company context: setup step, every run

The skill has no memory and no fixed place for company data. Context comes from wherever the user was launched, or from the user directly, and nothing is used without their say-so.

Run this after the entry gate and before the intake question:

1. Scan the launch directory (current working directory and its repo) for likely context. Look for CLAUDE.md, AGENTS.md, a context/ folder, files whose names mention brand, positioning, ICP, persona, voice, or competitors, and any earlier content-strategy-*.md or content-audit-*.md this skill produced. List filenames only. Don't read contents yet.
2. Ask one question. Show what was found (or say nothing relevant was found) and offer four choices: use these files; point me at a different folder or file; paste the context as text; start with no context. Wait.
3. Read only what the user approved. A pasted block counts as the context document for this run. "No context" means the intake answer is the whole picture; say so in the output document.
4. Earlier output is memory. If the user approves a previous content-strategy-*.md or content-audit-*.md, offer to continue from it (audit a built strategy, refresh a hub, re-score) rather than start over.

Optional extra: references/*-context.local.md packs, if any exist on this machine, are listed in step 2 alongside the scan results and follow the same rule: nothing is read until the user picks it. They never ship with the skill.

Entry gate

Offer text for the passive case:

I have a content-strategy skill that fits this. It either builds a documented SEO-driven strategy from scratch (7 parts, then a quarterly content hub of about 12 connected pieces), or audits what you already publish against the same 7 parts and ranks the gaps. Want me to run it? Build, audit, or not now?

"Build" or "audit" proceeds to the context setup step, then Step 0, with the mode pre-selected. "Not now" ends the skill's involvement. Hand the conversation back and don't raise it again unless the user does.

Step 0: Intake (always)

Ask this, in one message, nothing else:

Tell me about your content program today: what do you publish, for whom, and what is it supposed to do for the business?

Ask it even when context was approved and even when the user pre-selected a mode. Then route:

• Nothing real yet, or "starting from scratch" → Build flow.
• A running program with several of the seven parts in place → Audit flow.
• Content exists (even a lot of it) but none of the seven parts is in place: no mission, no research, no outside publishing → recommend Build and say why: there is nothing structural to score yet. The user may still pick Audit.
• Unclear → ask which they want: a strategy built, or the current program audited.

Data hooks (both flows)

If GA4 (analytics-mcp), Search Console (gsc), or DataForSEO tools are loaded, offer once, at part 1, to pull: conversion rate split by landing page type (blog vs service), top-linked pages, Domain Authority, phrase difficulty. If they aren't loaded or the user declines, ask for the numbers. If the user doesn't have them, record "unknown" once and reuse that answer wherever the number comes up again; don't re-ask. In the document, say why a number is unknown: "no analytics tools loaded this session" is different from "the user doesn't track this". Never make the numbers a precondition.

When the user can name roles but not people or sites ("two HR podcasts", "a trade site"), accept the role as a placeholder, record it, and make naming them a week-one task in the hub. Don't stall the flow on names.

Build flow

One question per message. Read the reference for a step before asking its question. Record each decision as you go; they become the document.

After step 7, read references/content-hub-template.md and assemble Quarter 1: twelve connected pieces, each with a stated purpose, on a twelve-week calendar.

Audit flow

1. Read references/audit-rubric.md. When scoring a part or writing its first fix, read that part's reference too (same files as the Build table).
2. For each of the seven parts, ask for evidence: URLs, examples, numbers. One part per message. Offer the data hooks at part 1 per the section above.
3. Score each part present / partial / missing against the rubric's observable criteria. Quote the evidence: a URL, a number, or the user's own sentence in quotation marks, labeled as which.
4. Compute the 1% test: how many of the four rare practices (documented mission, research-anchored, ongoing influencer collaboration, PR-focused) are true.
5. Rank the gaps by link-and-authority impact per the rubric, then adjust for what the user can realistically do this quarter.
6. Give one concrete first fix per gap.
7. Propose a next-quarter hub built only from the missing parts.

Output contract

Build document sections, in order:

1. Content mission (one sentence, plus audience / topics / benefit spelled out)
2. Sales page target: phrase, current DA, difficulty, gap, rank fixes, convert fixes
3. Original research: question, method, sample, effort estimate, promotion plan
4. Prospect questions captured, each mapped to an article title and a sales use
5. Visual repurposing plan
6. Influencer collaboration plan: formats chosen, named people or roles
7. Guest posting plan: target sites, pitch angles
8. Quarter 1 content hub: twelve slots with purpose and links between pieces, on a calendar
9. Timeline expectation and what to measure

Audit document sections, in order:

1. Scorecard: seven parts, score, evidence
2. The 1% test result
3. Gaps ranked, with one first fix each
4. Next-quarter hub from the missing parts

Default file names: content-strategy-[company].md (build) or content-audit-[company].md (audit), in the current directory. Confirm the path with the user before writing. Run /copy-deslop on the finished file.

Every claim in the document ties back to something the user said or a tool returned. Unknowns are labeled unknown, not estimated.

Quick reference

Common mistakes

• Starting with blog topics instead of the sales page. The page is the mousetrap; content is the cheese. Build the trap first.
• A mission everyone "knows" but nobody wrote down. Unwritten means undefined.
• A research idea too big to finish in a quarter. Scope to one headline finding.
• Guest posts that earn no link, or a link only in the bio when the research could have been cited in the body.
• Hub pieces that don't reference each other. The hub is a cluster, not a list.
• Reading context files because they were there. Found is not approved. Ask.
• Skipping the intake because context was loaded. Ask anyway.
• Starting work on a passive trigger. Offer, ask, wait.
• Filling unknown numbers with plausible ones. Label them unknown.