seo-geo-site-audit

Name: seo-geo-site-audit
Author: iklynow-hue

32stars0forks

How to install seo-geo-site-audit?

iklynow-hue/seo-geo-skills/seo-geo-site-audit

$npx -y skills add iklynow-hue/seo-geo-skills --skill seo-geo-site-audit

Installs into the current project.

Summary

Use when the user asks for an SEO audit, GEO audit, AI visibility review, technical content-readiness review, site-quality review, crawlability check, or asks Claude to audit, score, or grade a public website — even if they don't say the word "skill". The skill crawls a represent

Files · 1

View on GitHub

SKILL.md

1	# SEO GEO Site Audit
2
3	Turn a public website into a structured, evidence-based SEO + GEO audit.
4
5	## How to invoke
6
7	Just give the skill the target URL. It runs immediately with defaults. Override inline if needed.
8
9	\| Default \| Value \|
10	\|---\|---\|
11	\| Mode \| Light template audit (10 pages) \|
12	\| Output style \| Operator \|
13	\| Performance \| Local Lighthouse on the homepage (mobile + desktop) \|
14	\| HTML report \| On \|
15	\| Output language \| English \|
16
17	Defaults match the common case. Don't ask the user 5 questions. Run the wrapper and write the report.
18
19	## Inline overrides
20
21	The agent parses these from the user's prompt and passes the matching flags to the wrapper:
22
23	\| Phrasing \| Override \|
24	\|---\|---\|
25	\| `fast`, `quick check`, `1 page`, `homepage only` \| `--mode fast` (1 page) \|
26	\| `template`, `standard`, `full`, `50 pages` \| `--mode template` (50 pages) \|
27	\| `N pages` (1–50) \| `--max-pages N` \|
28	\| `boss style`, `executive summary` \| `--output-style boss` \|
29	\| `specialist`, `detailed`, `deep` \| `--output-style specialist` \|
30	\| `skip performance`, `skip lighthouse`, `no pagespeed` \| `--skip-pagespeed` \|
31	\| `no html`, `json only`, `skip html` \| drop `--html-report` \|
32	\| `in Chinese`, `中文`, `用中文` \| `--report-language chinese` \|
33	\| `in <language>`, `report in <language>` \| `--report-language <language>` \|
34
35	If the user inlines all preferences in the first message, parse and run. If the user gave only a URL, run with defaults silently.
36
37	## When to actually ask
38
39	Three narrow cases. Otherwise, never ask:
40
41	1. No URL. "Audit my site" without a URL — ask for the URL.
42	2. Language explicitly mentioned but ambiguous — e.g. user typed in Spanish but didn't say "in Spanish for the report". Ask once: "Final report in English or Spanish?"
43	3. Wrapper prerequisites missing. If `--auto-install-prereqs` would download Lightpanda nightly or Camoufox, mention the install + ask permission once before running. Default to `--skip-prereq-check` and the urllib fallback if the user says no.
44
45	A vague session-level "be quick" or "no clarifying questions" is the same signal as "use defaults" — just run.
46
47	## Default language stays English
48
49	If the user typed the request in Chinese, the report still defaults to English unless the user explicitly asks for a non-English report. Don't auto-translate based on prompt language.
50
51	Switch only on:
52	- `2` selected after a deliberate language question (case 2 above),
53	- a phrase like "in Chinese", "中文", "Spanish report", "report in German",
54	- the `--report-language` CLI flag.
55
56	## Architecture (read on demand)
57
58	Two evidence tracks per HTML page — a raw Googlebot-style HTTP baseline and a JS-rendered DOM — then comparison. Rendered fetching falls through Scrapling → Lightpanda → agent-browser → urllib. SPA recovery layer (Scrapling retry, scroll+wait, DOM route hints), domain-aware route guesser, sitemap-first fallback. Performance evidence comes from local Lighthouse via `scripts/run_lighthouse.mjs`.
59
60	Full rules — fetcher priority, recovery triggers, SPA detection thresholds, route guess templates, Lighthouse invocation — live in [`references/architecture.md`](references/architecture.md).
61
62	Two principles to keep in working memory:
63
64	- Raw vs rendered. If a signal appears only after JS, call it a JavaScript-dependency risk, not "Google cannot see it." If both raw and rendered are missing, it is a true missing signal.
65	- Assisted discovery. Pages reached only through `dom_route_hint` or `route_guess` are not search-discoverable. Don't present them as crawlable unless a raw `<a href>` or sitemap also exposes them.
66
67	## Guardrails
68
69	- Treat the crawl as a sample, not a full index.
70	- Default crawl cap is 10 pages (light mode). Maximum is 50.
71	- Stay on the same origin unless the user explicitly wants cross-domain review.
72	- Separate observed evidence from inference.
73	- Never imply access to Search Console, analytics, Ahrefs, SEMrush, or server logs unless the user actually provided them.
74	- If Lighthouse data is unavailable because of a node/npm/Chrome failure, complete the audit anyway and clearly label performance evidence as missing or partial.
75	- Never hardcode API keys, tokens, or secrets in this repo.
76
77	## Skill paths
78
79	The skill typically lives at `~/.claude/skills/seo-geo-site-audit/`. The wrapper script is `${SKILL_DIR}/scripts/audit-site`. Use `${SKILL_DIR}/...` or the user's actual install path; don't hardco

People who liked this also viewed

6 picks

Type

seo-geo-site-audit

32stars0forks

View on GitHub

How to install seo-geo-site-audit?

iklynow-hue/seo-geo-skills/seo-geo-site-audit

$npx -y skills add iklynow-hue/seo-geo-skills --skill seo-geo-site-audit

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# SEO GEO Site Audit
2
3	Turn a public website into a structured, evidence-based SEO + GEO audit.
4
5	## How to invoke
6
7	Just give the skill the target URL. It runs immediately with defaults. Override inline if needed.
8
9	\| Default \| Value \|
10	\|---\|---\|
11	\| Mode \| Light template audit (10 pages) \|
12	\| Output style \| Operator \|
13	\| Performance \| Local Lighthouse on the homepage (mobile + desktop) \|
14	\| HTML report \| On \|
15	\| Output language \| English \|
16
17	Defaults match the common case. Don't ask the user 5 questions. Run the wrapper and write the report.
18
19	## Inline overrides
20
21	The agent parses these from the user's prompt and passes the matching flags to the wrapper:
22
23	\| Phrasing \| Override \|
24	\|---\|---\|
25	\| `fast`, `quick check`, `1 page`, `homepage only` \| `--mode fast` (1 page) \|
26	\| `template`, `standard`, `full`, `50 pages` \| `--mode template` (50 pages) \|
27	\| `N pages` (1–50) \| `--max-pages N` \|
28	\| `boss style`, `executive summary` \| `--output-style boss` \|
29	\| `specialist`, `detailed`, `deep` \| `--output-style specialist` \|
30	\| `skip performance`, `skip lighthouse`, `no pagespeed` \| `--skip-pagespeed` \|
31	\| `no html`, `json only`, `skip html` \| drop `--html-report` \|
32	\| `in Chinese`, `中文`, `用中文` \| `--report-language chinese` \|
33	\| `in <language>`, `report in <language>` \| `--report-language <language>` \|
34
35	If the user inlines all preferences in the first message, parse and run. If the user gave only a URL, run with defaults silently.
36
37	## When to actually ask
38
39	Three narrow cases. Otherwise, never ask:
40
41	1. No URL. "Audit my site" without a URL — ask for the URL.
42	2. Language explicitly mentioned but ambiguous — e.g. user typed in Spanish but didn't say "in Spanish for the report". Ask once: "Final report in English or Spanish?"
43	3. Wrapper prerequisites missing. If `--auto-install-prereqs` would download Lightpanda nightly or Camoufox, mention the install + ask permission once before running. Default to `--skip-prereq-check` and the urllib fallback if the user says no.
44
45	A vague session-level "be quick" or "no clarifying questions" is the same signal as "use defaults" — just run.
46
47	## Default language stays English
48
49	If the user typed the request in Chinese, the report still defaults to English unless the user explicitly asks for a non-English report. Don't auto-translate based on prompt language.
50
51	Switch only on:
52	- `2` selected after a deliberate language question (case 2 above),
53	- a phrase like "in Chinese", "中文", "Spanish report", "report in German",
54	- the `--report-language` CLI flag.
55
56	## Architecture (read on demand)
57
58	Two evidence tracks per HTML page — a raw Googlebot-style HTTP baseline and a JS-rendered DOM — then comparison. Rendered fetching falls through Scrapling → Lightpanda → agent-browser → urllib. SPA recovery layer (Scrapling retry, scroll+wait, DOM route hints), domain-aware route guesser, sitemap-first fallback. Performance evidence comes from local Lighthouse via `scripts/run_lighthouse.mjs`.
59
60	Full rules — fetcher priority, recovery triggers, SPA detection thresholds, route guess templates, Lighthouse invocation — live in [`references/architecture.md`](references/architecture.md).
61
62	Two principles to keep in working memory:
63
64	- Raw vs rendered. If a signal appears only after JS, call it a JavaScript-dependency risk, not "Google cannot see it." If both raw and rendered are missing, it is a true missing signal.
65	- Assisted discovery. Pages reached only through `dom_route_hint` or `route_guess` are not search-discoverable. Don't present them as crawlable unless a raw `<a href>` or sitemap also exposes them.
66
67	## Guardrails
68
69	- Treat the crawl as a sample, not a full index.
70	- Default crawl cap is 10 pages (light mode). Maximum is 50.
71	- Stay on the same origin unless the user explicitly wants cross-domain review.
72	- Separate observed evidence from inference.
73	- Never imply access to Search Console, analytics, Ahrefs, SEMrush, or server logs unless the user actually provided them.
74	- If Lighthouse data is unavailable because of a node/npm/Chrome failure, complete the audit anyway and clearly label performance evidence as missing or partial.
75	- Never hardcode API keys, tokens, or secrets in this repo.
76
77	## Skill paths
78
79	The skill typically lives at `~/.claude/skills/seo-geo-site-audit/`. The wrapper script is `${SKILL_DIR}/scripts/audit-site`. Use `${SKILL_DIR}/...` or the user's actual install path; don't hardco

People who liked this also viewed

6 picks

Type