Implement ourdigital-presales-seo skill

SKILL.md orchestration (8 gated stages), references (rate_card.yaml,
findings_to_service rubric, competitor sets), findings.schema.json contract,
and scripts: kg_query.py (generalized KG examination), estimate.py
(findings→rate-card 견적 md/xlsx/json), build_deck.py (9-slide branded PPTX),
render_pdf.sh (Korean PDF via headless Chrome), plus client_brief.html template.

Validated on Sono Hotels & Resorts findings: estimate OD-2026-001
(23-47M KRW) and a 9-slide deck generated cleanly.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-05-27 23:10:53 +09:00
parent a55e77d1b0
commit ba88247496
10 changed files with 1034 additions and 0 deletions

View File

@@ -0,0 +1,89 @@
---
name: ourdigital-presales-seo
description: Standardized pre-sales SEO + Knowledge Graph diagnostic for a prospect domain that produces a technical/on-page scan, KG/entity analysis, consolidated brief, a rate-card-based cost estimate (견적), and an editable PPTX sales deck. Use for pre-sales prospecting, sales briefing prep, "pre-sales SEO audit", "프리세일즈 진단", "견적 + 제안 슬라이드", or when preparing for a prospect meeting. Public-data only (no client GSC/GA4 access required). Korean-first output.
---
# OurDigital Pre-sales SEO Workflow
Runs the full pre-sales diagnostic for **any prospect domain** as **step-by-step
confirmed stages**, ending in a cost estimate and a sales-briefing deck. Public-data
only — designed for prospects where you don't yet have GSC/GA4/GTM access.
Origin: standardizes the Sono Hotels & Resorts pre-sales diagnostic.
## Execution model — STEP BY STEP
Run one stage at a time. After each stage, **present the result and WAIT for the
user's go-ahead** before the next. Create a TodoWrite/Task item per stage. Stages 5
(estimate) and 6 (deliverables) are explicit review gates — never send without sign-off.
## Stage 0 — Scope & preflight
Collect (ask only for what you can't infer):
- `domain` (required); `brand_name` + `aliases` (앤/& / EN variants — infer from `<title>`/og first)
- `sub_brands`, `properties` (auto-extract from crawl URL patterns in Stage 1 if not given)
- `competitors` (else pick a set from `references/competitor_sets.md` by vertical)
- `market`/`language` (default South Korea / ko), `vertical`
- `output_dir` — default routing: if `~/Workspaces/<slug>-workspace/` exists →
`…/audits/<YYYY-MM-DD>-presales/`, else `~/Workspaces/seo-workspace/prospecting/<YYYY-MM-DD>-<prospect>/`
Preflight tool check (report missing + fallback): Firecrawl, DataForSEO, `GOOGLE_KG_API_KEY`,
headless Chrome, python-pptx. Create `data/` subfolder. Initialize `findings.json` (see `findings.schema.json`).
## Stage 1 — Discovery & crawl
- `WebFetch` robots.txt + `/sitemap.xml` + `/sitemap_index.xml` (note status; 500/404 = finding).
- `firecrawl_map` (limit high, includeSubdomains) → URL inventory; record `discoverable_urls`,
derive URL architecture, `estimated_pages`, and hygiene flags (`/test`, params, duplicate path schemes).
- Populate `findings.json.discovery`. → write scan §1.
## Stage 2 — Technical / on-page
- `firecrawl_scrape` (json) homepage + 1-2 property pages: JSON-LD `@type`s, Organization completeness
(sameAs/alternateName/address), meta/title/H1, hreflang set, robots/googlebot meta.
- `mcp__dfs-mcp__on_page_lighthouse` homepage (+ a property) → CWV.
- Populate `findings.json.technical`. → write `01_technical-onpage-scan.md`.
- **Honesty rule:** never claim "noindexed" if the page ranks — flag conflicting directives as "verify".
## Stage 3 — Knowledge Graph / entity
- `python scripts/kg_query.py --brand … --aliases … --parent … --legacy … --subbrands … --properties … --competitors … --out-dir <out>/data`
(needs `GOOGLE_KG_API_KEY`; run with sandbox disabled — read-only API GET).
- Live SERP panel verification: `mcp__dfs-mcp__serp_organic_live_advanced` (ko, South Korea) for brand,
a sub-brand, a property, and one competitor → record actual panel type/title, local packs, reseller leakage.
- Populate `findings.json.entity` (panel, name_split, legacy_contamination, sub-brand/property entity counts,
competitor_benchmark[], wikipedia). → write `02_knowledge-graph-entity.md` + `data/serp_panels_findings.md`.
## Stage 4 — Consolidated brief
- Synthesize, rank by severity, build the competitive-gap table. Populate `findings.json.findings[]`.
→ write `03_presales-opportunity-brief.md`.
## Stage 5 — Estimate (견적) — REVIEW GATE
- `python scripts/estimate.py --findings <out>/data/findings.json --rate-card references/rate_card.yaml --out-dir <out> --seq <N>`
- Produces `05_estimate_ko.md`, `05_estimate.xlsx`, `data/estimate.json`. Present the ranged 견적; get sign-off.
- Rules in `references/findings_to_service.md`; rates in `references/rate_card.yaml` (edit both together).
## Stage 6 — Deliverables — REVIEW GATE before send
- **Client PDF**: author the short brief HTML from `templates/client_brief.html` (fill the content; keep the CSS),
then `bash scripts/render_pdf.sh <brief>.html` → PDF. Verify Korean renders (Read the PDF).
- **Sales deck**: `python scripts/build_deck.py --findings <out>/data/findings.json --estimate <out>/data/estimate.json --out <out>/sales-deck.pptx`
- Sanitize the client-facing pieces: no internal pricing strategy beyond the 견적; tasteful competitor benchmark only.
## Stage 7 — Archive (standard)
Push the consolidated report to the OurDigital SEO Audit DB:
```
python <notion-writer path>/notion_writer.py \
--database 2c8581e5-8a1e-8035-880b-e38cefc2f3ef \
--title "<프로스펙트> SEO 사전진단 (Pre-sales) — <YYYY-MM-DD>" \
--properties '{"Target URL": "<domain>", "Audit Date": "<YYYY-MM-DD>", "Account Code": "<CODE>"}' \
--file <out>/03_presales-opportunity-brief.md
```
Property names are exact: `Target URL`, `Audit Date`, `Account Code`. **Do NOT set `Audit ID`**
(read-only formula). notion-writer path: `~/Project/our-claude-skills/custom-skills/32-notion-writer/code/scripts/notion_writer.py`.
## Tool gotchas (learned)
- OurSEO `crawl_website` **caps ~60 pages** regardless of `max_pages` — use Firecrawl `map` for inventory; report the discoverable count as a finding.
- Firecrawl `scrape` cache may return empty — re-scrape.
- KG API + headless Chrome + Notion push need network → run those Bash calls with the sandbox disabled.
- Korean PDF: headless Chrome uses system fonts (AppleSDGothicNeo on macOS). On Windows set deck font to Malgun Gothic in `build_deck.py`.
## Conventions
- Korean-first for all client-facing output; keep technical terms in English (SEO, CWV, Schema, hreflang).
- File names: `01_/02_/03_/05_` prefixes as above; raw artifacts in `data/`.
- Outputs go to the workspace (Stage 0 routing), **never** into this skills repo.