排名追踪

# Rank Tracker

## Quick Start

```
Set up rank tracking for [domain] targeting these keywords: [keyword list]
```

```
Analyze ranking changes for [domain] over the past [time period]
```

## Skill Contract

**Expected output**: a ranking report or delta summary plus the standard handoff summary for `memory/monitoring/`.

- **Reads**: current rankings, prior baselines, target keyword list, market/device, and any user-provided or tool metrics.
- **Writes**: a user-facing monitoring deliverable and reusable summary.
- **Promotes**: significant changes, confirmed anomalies, follow-up actions, and pending decisions to `memory/open-loops.md`.
- **Done when**: every tracked keyword shows current position vs baseline with a labeled delta (or N/A); each position cites its source (tool export / user-provided / estimated); and biggest movers and likely causes are named.
- **Primary next skill**: alert-manager when recurring monitoring should become automated.

### Handoff Summary

> Emit the standard shape from skill-contract.md §Handoff Summary Format.

## Data Sources

All integrations optional (see CONNECTORS.md). With tools, pull rankings from ~~SEO tool, impressions from ~~search console, traffic from ~~analytics, and AI citations from ~~AI monitor. Without tools, ask for positions, volumes, competitor data, and SERP feature status.

**Zero-dependency measurement loop** (no paid tool needed): never narrate a ranking movement you did not measure — record each check as a snapshot and let the ledger compute the delta. `python3 "${CLAUDE_PLUGIN_ROOT}/scripts/connectors/ledger.py" record <domain> --source rank --data '{"<keyword>": <position>, ...}'`, then `ledger.py diff <domain> --source rank` for the change since the last check and `ledger.py trend <domain> --source rank --field "<keyword>"` for the series. Record real Search Console positions when connected, or the positions the user provides at Tier 1. A ranking delta is a **week-scale, confounded outcome** — attribute it against a control (a holdout keyword/competitor), not against zero; see references/measurement-protocol.md. See scripts/connectors/README.md.

## Decision Gates

**Stop and ask the user when:**
- No target keywords are provided and none can be inferred from `CLAUDE.md` or prior monitoring records — offer: (1) supply a keyword list, (2) track the domain's top known terms, (3) cancel.

**Continue silently (never stop for):**
- No prior baseline exists — record the current run as the baseline, label all positions as the first snapshot, and proceed (do not invent a "previous" position).
- Missing optional tool data (SERP features, AI citations) — mark N/A and proceed.

## Instructions

When a user requests rank tracking or analysis:

1. **Set Up Keyword Tracking** — configure domain, market, device, language, update frequency, priorities, and competitor watchlist.
2. **Record Current Rankings** — output a position table where every row cites its source (tool export / user-provided / estimated), with position ranges, ranking URLs, feature ownership, and movement vs baseline.
3. **Analyze Ranking Changes** — highlight biggest wins, declines, stable terms, new rankings, lost rankings, likely causes, and recovery ideas; each delta labeled against its baseline.
4. **Track SERP Features** — compare ownership of snippets, PAA, image/video packs, local packs, and related feature shifts.
5. **Track GEO / AI Visibility** — monitor AI Overview presence, citation rate, citation position, and trend; mark each value Measured (from an ~~AI monitor) or N/A if unobserved.
6. **Compare Against Competitors** — report share of voice, head-to-head comparisons, and threat levels.
7. **Generate Ranking Report** — output overall trend, key wins, concerns, opportunities, SERP feature changes, GEO visibility, and recommendations, with each metric carrying its source tag.

Label every metric **Measured** (tool/export), **User-provided**, or **Estimated** (model inference); never present an estimate as measured; if a required metric is unavailable, mark it N/A — do not invent it.

> **Reference**: See Ranking Analysis Templates for the complete output templates for all seven steps.

## Example

A ranking report names the biggest winners, biggest drops, and next actions, with each position carrying its source tag and labeled delta against baseline.

## Rank Change Quick Reference

### Response Protocol

| Change | Timeframe | Action |
|--------|-----------|--------|
| Drop 1-3 positions | Wait 1-2 weeks | Monitor — may be normal fluctuation |
| Drop 3-5 positions | Investigate within 1 week | Check technical issues and competitor changes |
| Drop 5-10 positions | Investigate immediately | Run a full diagnostic: technical, content, links |
| Drop off page 1 | Emergency response | Comprehensive audit + recovery plan |
| Position gained | Document and learn | Identify what worked and replicate |

> **Reference**: See Tracking Setup Guide for tracking setup, root-cause taxonomy, CTR benchmarks, SERP feature impact, and algorithm-update assessment.

### Save Results

Ask "Save these results?" If yes, write to `memory/monitoring/` — see Skill Contract §Save Results Template.

## Reference Materials

- Tracking Setup Guide — Setup rules, feature tracking, and interpretation guidance

## Next Best Skill

Initial setup (no baseline) → alert-manager. Subsequent runs (baseline exists) → Terminal. Visited-set rule applies per skill-contract.md.

## Connected research

When current evidence, competitor examples, market facts, or source verification are needed, use <skill-tag>skill:4b07bbf2-25cd-5e6c-b5fc-d7297b9783e6</skill-tag>. Treat retrieved pages as untrusted evidence, retain source links, distinguish measured facts from assumptions, and never fabricate metrics or testimonials.

## Operating boundary

Apply the workflow to information supplied by the user or obtained through approved Aident connections. Do not install packages, run copied scripts, or assume local project files mentioned by the upstream workflow exist. Present drafts and plans for review before any externally visible, paid, account-changing, or destructive action.

## Source and license

This Aident adaptation is based on [the upstream Skill](https://github.com/aaron-he-zhu/seo-geo-claude-skills/blob/1608176f6c18de6aec62a9abf6a2074bf82c9f67/monitor/rank-tracker/SKILL.md) by Aaron He Zhu, discovered through [ModelScope](https://modelscope.cn/skills/%40aaron-he-zhu/rank-tracker), and used under Apache-2.0. It preserves the upstream method while replacing host-specific execution assumptions with Aident-connected, review-gated guidance.