seo-orchestrator
Orchestrates a full SEO + AI-search audit — acquires the site with the bundled scripts (crawl/snapshot), detects platform and vertical, runs the deterministic checks, dispatches the read-only specialist subagents in parallel with a dispatch envelope, merges their findings into a persisted report, and presents the two scores. Invoked by the `audit` and `geo` commands; not called directly.
Hainrixz · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add Hainrixz/claude-seo-ai --skill seo-orchestrator技能说明
name: seo-orchestrator
description: Orchestrates a full SEO + AI-search audit — acquires the site with the bundled scripts (crawl/snapshot), detects platform and vertical, runs the deterministic checks, dispatches the read-only specialist subagents in parallel with a dispatch envelope, merges their findings into a persisted report, and presents the two scores. Invoked by the audit and geo commands; not called directly.
user-invocable: false
allowed-tools: Read, Grep, Glob, WebFetch, Bash(node "${CLAUDE_PLUGIN_ROOT}/scripts/"), Bash(mkdir -p "${CLAUDE_PLUGIN_DATA}/runs/"), Bash(cat > "${CLAUDE_PLUGIN_DATA}/runs/*"), Agent
seo-orchestrator (Layer 2)
Coordinates the whole audit in three phases: detect → dispatch → synthesize. The scripts are the acquisition layer; the model only judges. Read the compact stdout summaries and the JSON files they name — never paste HTML into the conversation. Nothing is written to the user's project; every artifact lives under ${CLAUDE_PLUGIN_DATA}/runs/<host>/<run-id>/.
1. Detect — one call
- Resolve the target and the flags forwarded by the command (
--pages,--max,--render,--ua,--vertical,--environment,--feed,--out; default--out "${CLAUDE_PLUGIN_DATA}/runs"). - Run the whole deterministic pipeline in one command.
audit.mjsacquires the target (it callscrawl.mjsfor a URL,snapshot.mjsfor--pages 1or a local path, and skips acquisition entirely when handed a run directory), writesprofile.json(platform / framework / cms plugins / hosting / environment / capabilities / write_targets /vertical_hints, plus a vertical guess), runs the checks registry, and callsreport.mjsfor a first deterministic-only report:
node "${CLAUDE_PLUGIN_ROOT}/scripts/audit.mjs" <url|path> \
--out "${CLAUDE_PLUGIN_DATA}/runs" --checks deterministic --format json \
[--pages N] [--max N] [--render static|auto|js] [--ua <preset>] \
[--vertical <ids>] [--environment production|preview|staging|local] [--feed <path>]
Read only the JSON summary it prints: run_dir, mode, platform, vertical, coverage, tier, scores, findings, checks, report_json, report_md, findings_json, warnings. --render defaults to static here — pass --render auto when the target looks client-rendered. Exit 2 means the target could not be acquired; exit 3 only means a --fail-under / --fail-on-* gate tripped, which is a CI concern, not a failed audit.
3. Read <run_dir>/crawl.json (pages with slug/url/role, templates, sampling, warnings) and <run_dir>/profile.json.
4. Vertical: run seo-vertical-detect over the homepage parsed (<run_dir>/pages/<homepage-slug>.json) plus profile.vertical_hints. The script only guesses — profile.vertical.source is "inferred" unless --vertical was passed. If your reading adds or changes a vertical, re-run the deterministic pass over the same run (no re-crawl) so the conditional checks actually fire:
node "${CLAUDE_PLUGIN_ROOT}/scripts/audit.mjs" <run_dir> --checks deterministic --format json --vertical <ids>
references/routing.md maps vertical → conditional modules (M18 e-commerce, M19 local, M20 hreflang on multilingual).
5. The run now holds <run_dir>/findings.deterministic.json (the deterministic findings the agents must not re-emit) and <run_dir>/checks.json (checks run, errors, needs_api, manual_review, dropped).
2. Dispatch (parallel — one message, four Agent calls)
Spawn the four read-only specialists in one message so their verbose intermediate output stays isolated. Each prompt = the envelope block + that agent's module list. Agents never rely on ${…} substitution: pass absolute paths.
ENVELOPE
plugin_root: <absolute ${CLAUDE_PLUGIN_ROOT}>
run_dir: <absolute run dir>
pages: [{slug, url, role}, …] # <run_dir>/pages/<slug>.json (+ <slug>.html, + <slug>.rendered.html when rendered)
site: robots=<run_dir>/site/robots.json sitemaps=<run_dir>/site/sitemaps.json discovery=<run_dir>/site/discovery.json
vertical: {primary: <v>, also: [<v>…], multilingual: <bool>}
platform: <run_dir>/profile.json # one line: <platform>/<framework>, env=<kind>, head_owner=<…>
platform_cards: [<plugin_root>/references/platforms/<id>.md, …] # omit when the platform is unknown
modules: [<M-ids for this agent>]
deterministic_findings: <run_dir>/findings.deterministic.json — do not re-emit these ids; add model-judged findings only
return: JSON array only — findings per schema/finding.schema.json, no prose
| Agent | Modules |
|---|---|
technical-auditor | M1, M2 (+M3), M4, M7, M7b (mobile), M7c (headings), M8, M9, M10, M15, M17 — plus M20 (hreflang) when vertical.multilingual |
ai-search-geo-specialist | M6, M11, M12, M14, M21 (AI discovery & agent endpoints, weight 0), M22 (agent-readiness) |
content-eeat-analyst | M13, M16 |
schema-generator | M5 — plus M18 when ecommerce ∈ vertical, M19 when local-business ∈ vertical |
Subagents have no Write/Edit — the audit can never mutate files. If an agent returns prose around the array, keep only the array.
3. Synthesize
- Save each returned array verbatim to
<run_dir>/agents/<agent>.jsonvia a Bash heredoc (mkdir -p "<run_dir>/agents" && cat > "<run_dir>/agents/<agent>.json" <<'EOF' … EOF) — into the plugin data dir, never the project. This is the one Bash call outside thenodepattern; themkdir -p/cat >entries inallowed-toolspre-approve it for paths under${CLAUDE_PLUGIN_DATA}/runs/(TO-VERIFY in the smoke — a heredoc body containing;,|or&&may still prompt). node "${CLAUDE_PLUGIN_ROOT}/scripts/report.mjs" <run_dir> --merge "agents/*.json" --lang <en|es> [--vertical <ids>] [--environment <kind>] [--out-md <path>]— re-readsfindings.deterministic.json, merges the agent arrays (dedupe by id + normalized location, most severe wins;needs_api/ `m