perf: reduce context duplication in WebPerf skills (-16.6% lines)

.claude/skills/webperf-core-web-vitals/SKILL.md

@@ -25,7 +25,6 @@ JavaScript snippets for measuring web performance in Chrome DevTools. Execute wi
 - `scripts/LCP-Video-Candidate.js` — LCP Video Candidate
 - `scripts/LCP.js` — Largest Contentful Paint (LCP)
 
-Descriptions and thresholds: `references/snippets.md`
 
 ## Common Workflows
 
@@ -54,7 +53,7 @@ When LCP is slow or the user asks "debug LCP" or "why is LCP slow":
 When layout shifts are detected or the user asks "debug CLS" or "layout shift issues":
 
 1. **CLS.js** - Measure overall CLS score
-2. **Layout-Shift-Loading-and-Interaction.js** (from Interaction skill) - Separate loading vs interaction shifts
+2. **Layout-Shift-Loading-and-Interaction.js** _(pending — available in webperf-interaction skill)_
 3. Cross-reference with **webperf-loading** skill:
    - Find-Above-The-Fold-Lazy-Loaded-Images.js (lazy images causing shifts)
    - Fonts-Preloaded-Loaded-and-used-above-the-fold.js (font swap causing shifts)
@@ -63,11 +62,11 @@ When layout shifts are detected or the user asks "debug CLS" or "layout shift is
 
 When interactions feel slow or the user asks "debug INP" or "slow interactions":
 
-1. **INP.js** - Measure overall INP value
-2. **Interactions.js** (from Interaction skill) - List all interactions with timing
-3. **Input-Latency-Breakdown.js** (from Interaction skill) - Break down input delay, processing, presentation
-4. **Long-Animation-Frames.js** (from Interaction skill) - Identify blocking animation frames
-5. **Long-Animation-Frames-Script-Attribution.js** (from Interaction skill) - Find scripts causing delays
+1. **INP.js** - Measure overall INP value; call `getINP()` after interactions, `getINPDetails()` for full list
+2. **Interactions.js** _(pending — available in webperf-interaction skill)_
+3. **Input-Latency-Breakdown.js** _(pending — available in webperf-interaction skill)_
+4. **Long-Animation-Frames.js** _(pending — available in webperf-interaction skill)_
+5. **Long-Animation-Frames-Script-Attribution.js** _(pending — available in webperf-interaction skill)_
 
 ### Video as LCP Investigation
 
@@ -146,7 +145,7 @@ Use this decision tree to automatically run follow-up snippets based on results:
 
 ### After CLS.js
 
-- **If CLS > 0.1** → Run **webperf-interaction:Layout-Shift-Loading-and-Interaction.js** to separate causes
+- **If CLS > 0.1** → Run **webperf-interaction:Layout-Shift-Loading-and-Interaction.js** _(pending — available in webperf-interaction skill)_
 - **If CLS > 0.25 (poor)** → Run comprehensive shift investigation:
   1. **webperf-loading:Find-Above-The-Fold-Lazy-Loaded-Images.js** (images without dimensions)
   2. **webperf-loading:Fonts-Preloaded-Loaded-and-used-above-the-fold.js** (font loading strategy)
@@ -163,9 +162,20 @@ Use this decision tree to automatically run follow-up snippets based on results:
   3. **webperf-interaction:Long-Animation-Frames.js** (blocking frames)
   4. **webperf-interaction:Long-Animation-Frames-Script-Attribution.js** (culprit scripts)
 - **If specific interaction type is slow (e.g., keyboard)** → Focus analysis on that interaction type
+- **After INP data is collected** → call `getINPDetails()` for the full sorted interaction list (useful for identifying patterns across multiple slow interactions)
+
+### Error Recovery
+
+- **If any script returns `status: "error"`** → Check if the page has finished loading:
+  - If early in load: wait and re-run the script
+  - If page is an SPA: user may need to navigate to the target route first
+- **If LCP.js / LCP-Sub-Parts.js returns `status: "error"`** → Tell the user: "LCP data is not available yet. Please ensure the page has fully loaded, then run the analysis again."
+- **If INP.js `getINP()` returns `status: "error"`** → The `getDataFn: "getINP"` field signals the agent can retry after user interaction. Prompt the user to click, type, or scroll, then call `getINP()` again.
 
 ### Cross-Skill Triggers
 
+> **Context fork note:** This skill runs with `context: fork`. Cross-skill triggers below are **recommendations to report back to the parent agent**, not direct calls this subagent can execute. When a cross-skill trigger fires, tell the user which skill and script to run next. Scripts marked _(pending)_ are not yet available — skip them and note the limitation.
+
 These triggers recommend using snippets from other skills:
 
 #### From LCP to Loading Skill
@@ -223,6 +233,4 @@ When multiple CWV metrics are poor, prioritize investigation:
 ## References
 
 - `references/snippets.md` — Descriptions and thresholds for each script
-- `references/schema.md` — Return value schema for interpreting script output
-
-> Execute via `mcp__chrome-devtools__evaluate_script` → read with `mcp__chrome-devtools__get_console_message`.
+- `references/schema.md` — Return value schema for interpreting script output

.claude/skills/webperf-core-web-vitals/references/schema.md

@@ -186,6 +186,7 @@ Keep the existing `async () => {}` wrapper. Add a `return` statement with struct
 ```
 
 #### CLS
+Returns buffered CLS immediately and keeps tracking. Always call `getCLS()` after interactions to get an updated value.
 ```json
 {
   "script": "CLS",
@@ -194,9 +195,12 @@ Keep the existing `async () => {}` wrapper. Add a `return` statement with struct
   "value": 0.05,
   "unit": "score",
   "rating": "good",
-  "thresholds": { "good": 0.1, "needsImprovement": 0.25 }
+  "thresholds": { "good": 0.1, "needsImprovement": 0.25 },
+  "message": "CLS tracking active. Call getCLS() for updated value after page interactions.",
+  "getDataFn": "getCLS"
 }
 ```
+`getCLS()` returns the same shape with the latest accumulated value.
 
 #### INP (tracking)
 ```json
@@ -224,6 +228,15 @@ Keep the existing `async () => {}` wrapper. Add a `return` statement with struct
   }
 }
 ```
+If no interactions yet, `getINP()` returns `status: "error"` with `getDataFn: "getINP"` — retry after user interaction.
+
+`getINPDetails()` returns the full sorted interaction list (array of up to 15 entries). Use when `getINP()` shows poor INP and you need to identify patterns across multiple slow interactions:
+```json
+[
+  { "formattedName": "click → button.submit", "duration": 450, "startTime": 1200,
+    "phases": { "inputDelay": 120, "processingTime": 280, "presentationDelay": 50 } }
+]
+```
 
 #### LCP-Sub-Parts
 ```json