agentleFS
risukisurisukisu/marketing-skills/skills/Marketing-Ops/content-audit/SKILL.md

content-audit

Skill6 starsChanged 49 days agoMITView on GitHub

Diagnose how a website's content works as an inbound lead channel. Crawls the site, derives the offerings from the site as a whole (confirmed by the user), classifies every page by offering and AIDA stage (Attention, Interest, Desire, Action) with parallel subagents, scores value-proposition clarity, and finds where each offering's buyer journey breaks: missing or thin stages, pages with no onward step, content that is not found in search, traffic that does not convert. Optional Search Console, GA4, HubSpot and Google Ads exports, dropped into an ingest folder after the first run, add findability, traffic and first-touch leads. Python does every count. Output is one readable HTML report plus markdown and a page-level CSV, ending in a per-offering verdict (content, findability, conversion or distribution gap) and priorities handed to other skills. Use when the user says "/content-audit", "content audit", "content gap analysis", "AIDA audit", "journey gaps", "which offering lacks decision content", "does our content generate leads", or wants to know whether to invest in content, SEO, distribution or paid. Not the 7-part strategy audit (/content-strategy), not an internal-link audit (/internal-linking), not a technical SEO audit (/seo-audit).

26 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use asset-spec, create-control-manifest and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • lilangMaxSkill

    asset-spec

    Generate per-asset visual specifications and AI generation prompts from GDDs, level docs, or character profiles. Produces structured spec files and updates the master asset manifest. Run after art bible and GDD/level design are approved, before production begins.

    Also in 10 projects that use this

  • lilangMaxSkill

    create-control-manifest

    After architecture is complete, produces a flat actionable rules sheet for programmers — what you must do, what you must never do, per system and per layer. Extracted from all Accepted ADRs, technical preferences, and engine reference docs. More immediately actionable than ADRs (which explain why).

    Also in 10 projects that use this

  • +10 more from lilangMax

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 3 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 25 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask one question: use these files / point me at another folder or file / paste the context as text / start with none.

    line 64
  2. Distinctive:

    Read only what was approved.

    line 66
  3. Distinctive:

    Step 1: Intake question: Always ask it, once, even when the request seems to answer it: pre-fill what the user already said and let them …

    line 69
  4. Very specific:

    If Step 0 found a run folder with metrics.json, ask instead: continue that run (load new exports and rebuild) or start a new one.

    line 81
  5. Very specific:

    Step 2: Crawl: Run folder: content-audit-<host>-<YYYY-MM-DD> in the launch folder (<host> without www..

    line 84
  6. Very specific:

    Step 4: Offering map and the gate: Follow references/offering-map.md: write taxonomy.json with "confirmed": false, ask the confirmation …

    line 112
  7. Very specific:

    Step 9: Narrative and build: Write RUN/narrative.json per references/report-rules.md, run the clean-up pass it describes, then build:

    line 160

“Common” is measured: kept by at least 6 in 10 of the 25 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/content-audit — offering × AIDA content diagnosis

Crawl, map offerings, classify, score, measure, report. The scripts count; you judge offerings, stages and value propositions, and write the narrative. A run lives in one folder in the user's workspace and can be continued later with analytics exports.

ReferenceUse it for
references/offering-map.mdDeriving offerings, taxonomy.json, the confirmation question
references/classification.mdThe AIDA and offering rubric, and the subagent brief
references/value-proposition.mdThe six-criterion 0–2 score, valueprop.json
references/exports.mdWhich exports to request, the ingest folder, loader options
references/diagnosis-rules.mdCell statuses, verdict rules, thresholds, handoff skills
references/report-rules.mdnarrative.json, writing rules, build, final checklist

Requires Python 3.9+ on the PATH (python, python3 or py), standard library only. Scripts live in ~/.claude/skills/content-audit/scripts/; call them by full path from the launch folder so every output lands in the user's workspace, never in the skill folder (a public repository). Static HTML only: content rendered by JavaScript is not seen (the crawler warns).

Workflow

Copy this checklist into your response and tick items as you go:

Content audit:
- [ ] Step 0: Context intake (scan, list, ask, read only what is approved)
- [ ] Step 1: Intake question (one question; new run or continue a run)
- [ ] Step 2: Crawl (crawl_site.py)
- [ ] Step 3: Prepare (prepare.py) and read site_profile.md
- [ ] Step 4: Offering map, then the confirmation gate (taxonomy.json)
- [ ] Step 5: Classify in parallel subagents, validate (check_labels.py labels)
- [ ] Step 6: Value proposition scores (valueprop.json, check_labels.py valueprop)
- [ ] Step 7: Ingest folder (ingest.py init; load if files are already there)
- [ ] Step 8: Analyze (analyze.py)
- [ ] Step 9: Narrative, build, verify (build_report.py, report-rules.md checklist)
- [ ] Step 10: Announce the report and the ingest folder

Step 0: Context intake. The skill stores no company data and uses none without the user's say-so.

  1. Scan the launch directory (cwd and its repo) for CLAUDE.md, AGENTS.md, a context/ folder, files named for brand, positioning, offering, ICP, persona or marketing audit, and earlier content-audit-*/ run folders. List names only.
  2. Ask one question: use these files / point me at another folder or file / paste the context as text / start with none. Wait.
  3. Read only what was approved. A marketing audit or positioning doc informs the offering map and the narrative; it never overrides what the site says (differences between the two are findings).

Step 1: Intake question. Always ask it, once, even when the request seems to answer it: pre-fill what the user already said and let them correct it. Their own words about the site and its offerings are input for Step 4.

Which site, and anything I should know before I crawl it?

  1. Start URL (the homepage).
  2. Optional: offerings you already know of, brand terms, sections to skip (for example /de/ or /docs/), a different page cap (default 1,000).
  3. Analytics exports are not needed now: after the first report I create an ingest folder with exact export steps, and you can add Search Console, GA4, HubSpot or Google Ads files any time.

If Step 0 found a run folder with metrics.json, ask instead: continue that run (load new exports and rebuild) or start a new one. Continuing goes straight to Step 7 with load.

Step 2: Crawl. Run folder: content-audit-<host>-<YYYY-MM-DD> in the launch folder (<host> without www.; characters other than letters, digits, dots and hyphens become -).

python ~/.claude/skills/content-audit/scripts/crawl_site.py https://example.com/ --run-dir content-audit-example.com-2026-01-31

A large crawl takes time (about 1,000 pages in 10 to 15 minutes): run it in the background (run_in_background: true in Claude Code) and wait for it to finish, or give a foreground call the longest timeout for sites under about 400 pages. The crawler saves crawl.json every 50 pages; if a run stops, repeat the same command with --resume and it continues from there.

Options: --max-pages 1000, --delay 0.3, --exclude REGEX (repeatable; use for language folders and sections the user named), --no-sitemap, --urls list.txt (crawl only listed URLs), --resume. Exit code 2: the start URL failed; report it and ask for another. Read every warning: line; they go into the report. A JavaScript warning means the classification will rest on titles alone: tell the user before continuing.

Step 3: Prepare.

python ~/.claude/skills/content-audit/scripts/prepare.py --run-dir RUN

Excludes legal, utility, listing, careers, post-conversion and canonicalised duplicates (kept in the CSV with a reason), writes pages.json, site_profile.md and batches of 50 pages. Read site_profile.md in full.

Step 4: Offering map and the gate. Follow references/offering-map.md: write taxonomy.json with "confirmed": false, ask the confirmation question, apply the answer, set "confirmed": true, then:

python ~/.claude/skills/content-audit/scripts/check_labels.py --run-dir RUN taxonomy

Nothing is classified before this passes. This is the run's second and last question.

Step 5: Classify. One subagent per batch file, brief from references/classification.md ("Subagent brief"). Send up to 10 in one message, then the next wave. Subagents write RUN/labels/batch_NN.json and reply with one line; do not ask them for summaries. Then:

python ~/.claude/skills/content-audit/scripts/check_labels.py --run-dir RUN labels

It prints every problem and rerun: … batch numbers. Re-dispatch those batches with the problem lines added to the brief (at most twice); a batch that still fails, classify yourself. Without a subagent tool, classify batch by batch yourself, writing the same files.

Step 6: Value proposition. Score the homepage and each offering's main page per references/value-proposition.md, write RUN/valueprop.json, then run check_labels.py --run-dir RUN valueprop and fix what it lists.

Step 7: Ingest folder.

python ~/.claude/skills/content-audit/scripts/ingest.py --run-dir RUN init
python ~/.claude/skills/content-audit/scripts/ingest.py --run-dir RUN load    # only if files are there

init writes RUN/ingest/README.md with the export steps. load prints each file's detected kind, the columns used and the date window. Exit code 3 means an unrecognised file: show its columns to the user and ask what they are; never guess a mapping (references/exports.md lists the override flags). Warnings about truncated exports go into the report.

Step 8: Analyze.

python ~/.claude/skills/content-audit/scripts/analyze.py --run-dir RUN

Prints the offering × AIDA grid with statuses and primary gaps, and warns when an export matches under 70% of crawled pages (check the host or the crawl cap before writing anything). Read metrics.json for the narrative: per_offering (verdicts with evidence), matrix, icp_fit, freshness, brand_split, valueprop, duplicates, striking_distance, layers.

Step 9: Narrative and build. Write RUN/narrative.json per references/report-rules.md, run the clean-up pass it describes, then build:

python ~/.claude/skills/content-audit/scripts/build_report.py --run-dir RUN

Fix every unverified number and every style: line, rebuild, then tick the checklist in references/report-rules.md.

Step 10: Announce. Verdict and top priority in a few sentences, then the HTML path with the 📄 marker, the markdown and CSV paths. After a site-only run, name the ingest folder and what each export would add; after a Phase 2 run, say what changed from the site-only read.

When things go wrong

SituationDo this
Crawler exit 2Report the status; ask for another start URL (often www. vs bare domain)
Crawl stopped or timed outRerun the same command with --resume; it keeps the pages saved so far
Markup warning ("almost every link as navigation")Nothing to do: body links were separated by repetition across pages; mention it under limits
JavaScript warningSay so; offer --urls with a sitemap list, or continue on titles and headings and mark confidence low
Several languagesAudit one: rerun with --exclude for the other language folders
Page cap reachedSay which sections were covered; offer a rerun with a higher cap or excluded sections
A batch keeps failing validationClassify it yourself; note it in limits
Export not recognised (exit 3)Show its columns, ask the user, rerun load with --source/--column
Export window unknownAsk the user for the dates; rerun load with --window
Under 70% of an export matchesCheck host, language folder and cap before analysing; report the match share either way
An offering has no pagesKeep it; every cell shows missing, which is the finding

Scope

Diagnoses and points at fixes; does not write content, change the site, or plan the calendar. The priorities hand work to /content-strategy, /cro, /internal-linking, /seo-* and the other skills named in references/diagnosis-rules.md. Rerun the skill later on the same site to compare against an earlier run folder.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.