performance-monitor Checked

• **Reads**: prior baselines and current performance data. report reads current + prior-period metrics across traffic/rankings/authority/content, report audience, and date range. alert reads baselines, critical keywords/metrics to watch, normal volatility, and delivery preferences. Plus any user-provided or tool data.
• **Writes**: a user-facing monitoring deliverable plus a reusable summary storable under memory/monitoring/.
• **Promotes**: significant changes, confirmed anomalies, durable thresholds, and follow-up actions; pending decisions go to memory/open-loops.md (never directly to decisions.md).
• **Done when**:
• report: each in-scope section (traffic, rankings, GEO, authority, backlinks, content) is present or marked "Not yet evaluated"; every metric is source-, time-, window-, and target-bound and compared to the prior period/control; each changed page records the current page/content/change and measurement-contract refs/hashes; recommendations carry owner, priority, and expected impact.
• alert: each chosen alert category has a named trigger, threshold, and priority; a Critical/High/Medium/Low response plan and delivery routing are defined; thresholds are tuned to the metric's stated normal volatility.
• **Primary next skill**: see Next Best Skill (#next-best-skill) — the two modes hand off to each other.

### Handoff Summary

> Emit the standard shape from skill-contract.md §Handoff Summary Format (../../../references/skill-contract.md). Name the active mode (report / alert) in the Objective line.

Data Sources

All integrations optional (see CONNECTORS.md (../../../CONNECTORS.md)). Tier 1 (keyless) works for both modes; keyed tools are opt-in Tier 2/3.

• **report**: with tools connected, aggregates traffic from ~~analytics, search data from ~~search console, rankings/backlinks from ~~SEO tool, and AI visibility from ~~AI monitor. Without tools, ask the user for analytics exports, Search Console data, ranking data, and KPIs.
• **alert**: with tools, monitor real-time feeds from ~~SEO tool, ~~search console, and ~~web crawler. Without tools, ask for baselines, critical keywords, delivery preferences, and historical data.

**Zero-dependency measurement loop** (both modes): every reported change or fired alert should come from a computed delta, not an eyeballed estimate. Store each period's KPIs and let the ledger compute movement: python3 "${CLAUDE_PLUGIN_ROOT}/scripts/connectors/ledger.py" record <domain> --source monitor --data '{"sessions": ..., "clicks": ..., ...}', then ledger.py diff <domain> --source monitor for the period delta and ledger.py trend <domain> --source monitor --field <kpi> for the trend line. Label every figure Measured / User-provided / Estimated, and attribute outcome movement against a control rather than to the latest change — see references/measurement-protocol.md (../../../references/measurement-protocol.md). See scripts/connectors/README.md (../../../scripts/connectors/README.md).

Decision Gates

**Stop and ask the user when:**

• **Mode ambiguous** — the request fits neither past-tense ("report on last month") nor future-tense ("tell me when X drops"). Offer: (1) report a snapshot now, (2) alert for ongoing monitoring, (3) both.
• **report**: no reporting period or comparison period can be determined and none is in context — offer (1) last 30 days vs prior 30, (2) last calendar month vs prior month, (3) a custom range. A period comparison is required, not optional.
• **alert**: no baseline or normal-volatility reference exists for the metrics to watch — thresholds would be arbitrary. Offer (1) supply recent baseline data, (2) use the Alert Threshold Quick Reference defaults and label them Estimated, (3) cancel.

**Continue silently (never stop for):**

• A report section's source data is missing — mark that section "Not yet evaluated" and proceed; do not fabricate the metric.
• Audience not stated (report) — default to the executive template and note the assumption.
• Delivery channel not stated (alert) — default to the channel in context (or email) and note it.
• An alert category the user did not mention — leave it unconfigured; do not add alerts they did not request.

Instructions

**Step 0 — Select mode.** Read --mode if given. Otherwise infer: past-tense / "how did we do" / "月报" → report; future-tense / "tell me when" / "预警" → alert. If neither fits, use the mode-ambiguous gate above.

### Mode: report

Use Report Output Templates (references/report-output-templates.md) and cover:

1. **Define Report Parameters** — domain, period, comparison period, report type, audience, focus areas, data freshness.

2. **Executive Summary** — overall rating, wins, watch areas, required actions, metrics-at-a-glance (traffic, rankings, conversions, DA/authority, AI citations), and SEO ROI; tag each metric Measured / User-provided / Estimated.

3. **Organic Traffic** — sessions, users, pageviews, engagement/bounce, trend, source/device split, top pages, each figure source-tagged.

4. **Keyword Rankings** — position ranges, distribution change, top improvements/declines, SERP features. For raw position-by-position deltas, defer to rank-tracker (../rank-tracker/SKILL.md) rather than recomputing here.

5. **GEO / AI Performance** — AI citation overview, citations by topic, GEO wins, optimization opportunities.

6. **Domain Authority (CITE)** — include CITE dimension scores and veto status **when a gate has already produced them**; otherwise mark "Not yet evaluated." Do not compute CITE here.

7. **Co