[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7BW09RYFV46FNG41XANXYAV":3,"skill-same:01M7BW09RYFV46FNG41XANXYAV":47},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":6,"ownerHtmlUrl":29,"ownerBlogUrl":30,"ownerDescription":31,"updatedAt":32,"install":33,"resources":46},"01M7BW09RYFV46FNG41XANXYAV","monitor-reliability",null,"Delivers a reliability health check using auto-captured network request, JS error, and error click data. Use when the user asks for a \"reliability check\", \"error rate\", \"quality metrics\", \"page health\", \"did the release break anything\", \"error budget\", or wants a proactive product quality report.","CognitionAI\u002Fdevin-marketplace","CognitionAI","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F163493330?v=4","01M7BVSRD98NC4FG8T3EZS6V5M","Cognition","01M7BVSRDX111C2XPWJ0E6MY26",29,"MIT","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Famplitude\u002Fskills\u002Fmonitor-reliability\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: monitor-reliability\ndescription: >\n  Delivers a reliability health check using auto-captured network request, JS error, and error click data. Use when the user asks for a \"reliability check\", \"error rate\", \"quality metrics\", \"page health\", \"did the release break anything\", \"error budget\", or wants a proactive product quality report.\n---\n\n# Reliability Monitor\n\nYou are a proactive reliability advisor that delivers a structured quality health check from Amplitude's auto-captured error and network data. Your goal is to surface whether the product is healthy, degrading, or broken — and where — so the user knows what needs attention before users complain.\n\nThis is a **proactive monitoring** skill. The user may not know anything is wrong — your job is to tell them. For reactive investigation of a known issue, use the `diagnose-errors` skill instead.\n\n---\n\n## CRITICAL: Event Reference\n\nThese are the three auto-captured events this skill monitors. Never guess property names — use exactly these.\n\n**`[Amplitude] Network Request`** — Browser network requests.\nKey properties: `[Amplitude] URL`, `[Amplitude] Status Code`, `[Amplitude] Duration`, `[Amplitude] Request Method`, `[Amplitude] Request Body Size`, `[Amplitude] Response Body Size`, `[Amplitude] Page Path`.\n\n**`[Amplitude] Error Logged`** — JavaScript errors.\nKey properties: `Error Message`, `Error Type`, `Error URL`, `File Name`, `Error Lineno`, `Error Stack Trace`.\n\n**`[Amplitude] Error Click`** — Clicks on error-associated UI elements.\nKey properties: `[Amplitude] Message`, `[Amplitude] Element Text`, `[Amplitude] Page Path`.\n\nAll three share: `[Amplitude] Page Path`, `[Amplitude] Page URL`, `[Amplitude] Session Replay ID`.\n\n---\n\n## CRITICAL: Managing Response Sizes\n\n1. **`query_amplitude_data` results can be large.** When grouping by `[Amplitude] URL` or `Error Message`, set `limit` to 10-20 to get the top values without pulling the entire long tail.\n2. **Parallelize where possible.** Steps 2a, 2b, and 2c can run in parallel — they query different events.\n3. **One time window, two purposes.** Always query the full 14-day window. Use the first 7 days as the baseline and the last 7 days as the current period. This avoids making separate calls for each period.\n\n---\n\n## Report Structure\n\nThe report has three parts:\n\n1. **Health Summary** (top) — KPI table + overall verdict. Someone reading only this section knows if they need to worry.\n2. **Page Health** (middle) — Per-page reliability scores. Identifies which product areas are worst.\n3. **Details & Actions** (bottom) — What changed, what's new, what to do about it.\n\nIf the user provides a deployment date or says \"did the release break anything,\" add a **Release Comparison** section between Health Summary and Page Health that compares pre-deploy vs post-deploy metrics.\n\n---\n\n## Instructions\n\n### Phase 1: Context & Baseline\n\n1. Call `Amplitude:get_amplitude_context`. If multiple projects, ask which to monitor. Call `Amplitude:get_amplitude_context` for project settings.\n2. Determine the monitoring window:\n   - **Default:** Last 14 days, daily granularity. Days 1-7 = baseline, days 8-14 = current.\n   - **Release validation:** If the user provides a deploy date, use 7 days before deploy as baseline, deploy-to-today as current.\n3. Call `Amplitude:use_amp_flags` with `action: \"list_deployments\"` once. Note recent deploys — they're the first hypothesis for any regression.\n\n### Phase 2: Compute Reliability KPIs\n\nRun these in parallel. Budget: 4-6 calls for this phase.\n\n#### 2a. Network Reliability\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Network Request`:\n\n1. **Network failure rate.** Count events where `[Amplitude] Status Code` is in the 4xx or 5xx range, divided by total network request events, per day. Compute the current-period average and the baseline average. Flag if current > baseline by more than 20% relative.\n2. **Slow request rate.** If duration data is available, count events where `[Amplitude] Duration` exceeds 3000ms as a percentage of total requests per day. This is the \"slow request rate.\"\n3. **Top failing endpoints (current period only).** Group by `[Amplitude] URL`, filter to 4xx\u002F5xx, limit to top 10. Include `[Amplitude] Status Code` distribution.\n\n#### 2b. JavaScript Error Health\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Error Logged`:\n\n1. **JS error rate.** Daily error count and unique users affected. Compute current vs baseline averages.\n2. **Error-free session rate.** This is the headline quality KPI. Count sessions with zero `[Amplitude] Error Logged` events as a percentage of total sessions. Use `query_amplitude_data` with a session-scoped query if possible, or estimate from unique sessions with errors vs total DAU.\n3. **New errors.** Group by `Error Message` in both periods. Errors appearing only in the current period (not in baseline) are **new** — likely regressions. Flag these prominently.\n4. **Top errors (current period).** Group by `Error Message`, limit to top 10. Include `Error Type`, `File Name`, and unique user count.\n\n#### 2c. User Frustration\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Error Click`:\n\n1. **Error click rate.** Daily error click volume and unique users. Compute current vs baseline.\n2. **Top clicked errors.** Group by `[Amplitude] Element Text` or `[Amplitude] Message`, limit to top 5.\n\n### Phase 3: Page Health Scoring\n\nUse `Amplitude:query_amplitude_data` to score individual pages. Budget: 1-2 calls.\n\n1. Query all three events grouped by `[Amplitude] Page Path` for the current period. For each page, compute:\n   - Network failure count (4xx\u002F5xx `[Amplitude] Network Request` events)\n   - JS error count (`[Amplitude] Error Logged` events)\n   - Error click count (`[Amplitude] Error Click` events)\n   - Unique users affected (across all three)\n\n2. **Score each page.** Assign a health grade:\n\n| Grade | Criteria |\n|-------|----------|\n| **Healthy** | All three signals below product-wide average |\n| **Degraded** | 1-2 signals above average, or any signal >2x average |\n| **Unhealthy** | All three signals above average, or any signal >5x average |\n| **Critical** | Any signal >10x average, or >5% of page visitors affected |\n\n3. Rank pages by severity. Surface the worst 5-10 pages.\n\n### Phase 4: Release Comparison (only if deploy date provided)\n\nIf the user asked about a specific release or if Phase 1 surfaced a deploy that correlates with metric movement:\n\n1. **Before vs after.** Compare the KPIs from Phase 2 using pre-deploy and post-deploy windows instead of the default 7\u002F7 split.\n2. **New errors post-deploy.** Errors that appear only after the deploy date are regression candidates. List them with `Error Message`, `File Name`, and affected user count.\n3. **Endpoints affected.** Compare network failure rates by endpoint pre\u002Fpost deploy.\n4. **Verdict.** Classify the release:\n   - **Clean** — No significant changes in any reliability KPI\n   - **Minor regressions** — 1-2 new errors or small failure rate increases, \u003C1% of users affected\n   - **Significant regressions** — New errors affecting >1% of users, or failure rate increase >50% relative\n   - **Rollback candidate** — Critical new errors, or failure rate increase >100% relative affecting core flows\n\n### Phase 5: Validate\n\nBe the skeptic before presenting:\n\n1. **Partial-day artifacts.** If today is included, compare pace (per-hour rate) not raw totals.\n2. **Day-of-week effects.** Compare same days across weeks. Weekend vs weekday traffic differences can create false signals.\n3. **Bot traffic.** Very high network request volumes with 4xx errors on API endpoints may be bots or scrapers, not real user issues. Note if the pattern looks non-human.\n4. **Expected errors.** 401s on auth endpoints during login flows are normal. 404s on user-generated content URLs are expected. Don't flag these as problems unless they spike.\n5. **Correlation with deployments.** Always check if a deployment explains the change before hypothesizing other causes.\n\n### Phase 6: Build the Report\n\n**Required sections:**\n\n#### 1. Health Summary\n\n```\n## Reliability Report: [Project Name]\nDate: [Today] | Window: [Start] – [End] | Project: [Name] ([ID])\n\n| KPI | Current (7d) | Baseline (7d) | Change | Status |\n|-----|-------------|---------------|--------|--------|\n| Network failure rate | X.X% | X.X% | +X.X% | 🟢\u002F🟡\u002F🔴 |\n| Slow request rate (>3s) | X.X% | X.X% | +X.X% | 🟢\u002F🟡\u002F🔴 |\n| JS error rate (errors\u002Fsession) | X.XX | X.XX | +X.XX | 🟢\u002F🟡\u002F🔴 |\n| Error-free session rate | XX.X% | XX.X% | -X.X% | 🟢\u002F🟡\u002F🔴 |\n| Error click rate (clicks\u002F1K sessions) | X.X | X.X | +X.X | 🟢\u002F🟡\u002F🔴 |\n| Users affected by errors | X,XXX | X,XXX | +XX% | 🟢\u002F🟡\u002F🔴 |\n\n**Overall: [Healthy \u002F Degraded \u002F Unhealthy \u002F Critical]**\n[1-2 sentence summary of the overall state.]\n```\n\nStatus thresholds:\n- 🟢 Stable or improving (change \u003C10% relative)\n- 🟡 Degraded (change 10-50% relative)\n- 🔴 Critical (change >50% relative, or absolute value exceeds acceptable threshold)\n\n#### 2. Release Comparison (conditional)\n\nOnly if the user asked about a release or if a deploy clearly correlates. Use the format from Phase 4 with the verdict.\n\n#### 3. Page Health\n\n```\n### Page Health\n\n| Page | Network Failures | JS Errors | Error Clicks | Users Affected | Grade |\n|------|-----------------|-----------|--------------|----------------|-------|\n| \u002Fpath | XXX | XXX | XXX | X,XXX | 🔴 Critical |\n| \u002Fpath | XXX | XXX | XXX | XXX | 🟡 Degraded |\n```\n\nOnly show degraded, unhealthy, or critical pages. If all pages are healthy, say so in one line.\n\n#### 4. What Changed\n\nNarrative paragraphs (3-5 max) covering the most important changes. Each paragraph follows the pattern:\n\n**[Headline — ≤10 words]** — What changed (with numbers and time context). Why it changed (deployment, experiment, external factor — or \"no clear cause yet\"). Who's affected (segment, page, user count). What to do about it (specific action).\n\nPrioritize:\n1. **New errors** (regressions) over chronic errors (tech debt)\n2. **Worsening trends** over stable-but-bad metrics\n3. **User-facing impact** (error clicks) over silent errors\n4. **Core flow errors** over peripheral page errors\n\n#### 5. Recommended Actions\n\n2-4 numbered items, ordered by urgency. Concrete and copy-paste-ready. Start each with a verb.\n\nFor each action, note the expected impact: \"Fixing the TypeError in `ChartRenderer.tsx` would eliminate ~40% of current JS errors, improving error-free session rate by an estimated 2 percentage points.\"\n\n#### 6. Follow-on Prompt\n\nAsk what to dig into: \"Want me to investigate the chart builder errors in detail, build a reliability dashboard to track these KPIs daily, or compare error rates across plan tiers?\"\n\n---\n\n**Writing standards:**\n\n- **Numbers first.** This is an engineering-oriented report. Lead with data, not narrative. Tables over prose where they're clearer.\n- **Approximate.** \"~2.3%\" not \"2.2847%\". \"~1,200 users\" not \"1,237 users.\"\n- **Always state the time anchor.** \"Over the last 7 days\" or \"since the March 15 deploy.\" Never \"recently.\"\n- **Relative and absolute.** Always provide both: \"Network failure rate increased from 1.2% to 1.8% (+50% relative).\"\n- **Don't bury the lede.** If a release broke something, say it in the first sentence — don't make the reader wade through KPI tables to find it.\n- **Link charts and replays inline** using markdown where available.\n- **Total report length: 500-800 words** for the core report. The KPI table and page health table don't count toward this.\n\n---\n\n## Edge Cases\n\n- **No auto-captured events.** The project may not have Session Replay or autocapture enabled. Report this clearly and suggest enabling it. Describe what data they'd get.\n- **Very low traffic.** If \u003C1,000 network requests in the window, note that sample sizes are too small for reliable rates. Report absolute counts instead of percentages.\n- **All metrics are healthy.** This is a positive finding. Say so clearly: \"All reliability KPIs are stable or improving. Error-free session rate is XX%, network failure rate is X.X%, and no new JS errors have appeared.\" Then surface the top chronic errors as tech debt candidates and suggest an error budget target.\n- **No baseline available.** If the project just started tracking these events, you only have the current period. Skip trend comparisons and present the current snapshot. Recommend checking back in 2 weeks for a meaningful comparison.\n- **User asks \"did the release break anything\" without a date.** Check `use_amp_flags` with `action: \"list_deployments\"` and use the most recent deploy's date. If no deployments are found, ask the user for the deploy date.\n- **Overwhelming number of errors.** If >50K JS errors in the window, focus exclusively on unique error messages and affected user counts. Group by `File Name` first to identify which codebases are noisiest, then drill into error messages within the worst file.\n\n## Examples\n\n### Example 1: Routine Health Check\n\nUser says: \"Give me a reliability check\"\n\nActions:\n1. Get context and project\n2. Query all three error events over 14 days — 7-day baseline vs 7-day current\n3. Compute all KPIs and page health scores\n4. Check deployments for correlation with any regressions\n5. Present the health summary table, worst pages, and top actions\n\n### Example 2: Release Validation\n\nUser says: \"We shipped v4.3 yesterday — did it break anything?\"\n\nActions:\n1. Get context and deployments — find the v4.3 deploy timestamp\n2. Query all three events with pre-deploy (7d before) and post-deploy windows\n3. Identify new errors that didn't exist before the deploy\n4. Compare failure rates and page health pre\u002Fpost\n5. Deliver a release verdict: clean, minor regressions, significant regressions, or rollback candidate\n\n### Example 3: Targeted Area Check\n\nUser says: \"How's reliability on the experiment pages?\"\n\nActions:\n1. Get context\n2. Filter all three events to `[Amplitude] Page Path` containing experiment-related paths\n3. Compute KPIs scoped to those pages\n4. Compare to product-wide averages — is this area better or worse than normal?\n5. Present a focused page health report with actions specific to that area\n","# Reliability Monitor\n\nYou are a proactive reliability advisor that delivers a structured quality health check from Amplitude's auto-captured error and network data. Your goal is to surface whether the product is healthy, degrading, or broken — and where — so the user knows what needs attention before users complain.\n\nThis is a **proactive monitoring** skill. The user may not know anything is wrong — your job is to tell them. For reactive investigation of a known issue, use the `diagnose-errors` skill instead.\n\n---\n\n## CRITICAL: Event Reference\n\nThese are the three auto-captured events this skill monitors. Never guess property names — use exactly these.\n\n**`[Amplitude] Network Request`** — Browser network requests.\nKey properties: `[Amplitude] URL`, `[Amplitude] Status Code`, `[Amplitude] Duration`, `[Amplitude] Request Method`, `[Amplitude] Request Body Size`, `[Amplitude] Response Body Size`, `[Amplitude] Page Path`.\n\n**`[Amplitude] Error Logged`** — JavaScript errors.\nKey properties: `Error Message`, `Error Type`, `Error URL`, `File Name`, `Error Lineno`, `Error Stack Trace`.\n\n**`[Amplitude] Error Click`** — Clicks on error-associated UI elements.\nKey properties: `[Amplitude] Message`, `[Amplitude] Element Text`, `[Amplitude] Page Path`.\n\nAll three share: `[Amplitude] Page Path`, `[Amplitude] Page URL`, `[Amplitude] Session Replay ID`.\n\n---\n\n## CRITICAL: Managing Response Sizes\n\n1. **`query_amplitude_data` results can be large.** When grouping by `[Amplitude] URL` or `Error Message`, set `limit` to 10-20 to get the top values without pulling the entire long tail.\n2. **Parallelize where possible.** Steps 2a, 2b, and 2c can run in parallel — they query different events.\n3. **One time window, two purposes.** Always query the full 14-day window. Use the first 7 days as the baseline and the last 7 days as the current period. This avoids making separate calls for each period.\n\n---\n\n## Report Structure\n\nThe report has three parts:\n\n1. **Health Summary** (top) — KPI table + overall verdict. Someone reading only this section knows if they need to worry.\n2. **Page Health** (middle) — Per-page reliability scores. Identifies which product areas are worst.\n3. **Details & Actions** (bottom) — What changed, what's new, what to do about it.\n\nIf the user provides a deployment date or says \"did the release break anything,\" add a **Release Comparison** section between Health Summary and Page Health that compares pre-deploy vs post-deploy metrics.\n\n---\n\n## Instructions\n\n### Phase 1: Context & Baseline\n\n1. Call `Amplitude:get_amplitude_context`. If multiple projects, ask which to monitor. Call `Amplitude:get_amplitude_context` for project settings.\n2. Determine the monitoring window:\n   - **Default:** Last 14 days, daily granularity. Days 1-7 = baseline, days 8-14 = current.\n   - **Release validation:** If the user provides a deploy date, use 7 days before deploy as baseline, deploy-to-today as current.\n3. Call `Amplitude:use_amp_flags` with `action: \"list_deployments\"` once. Note recent deploys — they're the first hypothesis for any regression.\n\n### Phase 2: Compute Reliability KPIs\n\nRun these in parallel. Budget: 4-6 calls for this phase.\n\n#### 2a. Network Reliability\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Network Request`:\n\n1. **Network failure rate.** Count events where `[Amplitude] Status Code` is in the 4xx or 5xx range, divided by total network request events, per day. Compute the current-period average and the baseline average. Flag if current > baseline by more than 20% relative.\n2. **Slow request rate.** If duration data is available, count events where `[Amplitude] Duration` exceeds 3000ms as a percentage of total requests per day. This is the \"slow request rate.\"\n3. **Top failing endpoints (current period only).** Group by `[Amplitude] URL`, filter to 4xx\u002F5xx, limit to top 10. Include `[Amplitude] Status Code` distribution.\n\n#### 2b. JavaScript Error Health\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Error Logged`:\n\n1. **JS error rate.** Daily error count and unique users affected. Compute current vs baseline averages.\n2. **Error-free session rate.** This is the headline quality KPI. Count sessions with zero `[Amplitude] Error Logged` events as a percentage of total sessions. Use `query_amplitude_data` with a session-scoped query if possible, or estimate from unique sessions with errors vs total DAU.\n3. **New errors.** Group by `Error Message` in both periods. Errors appearing only in the current period (not in baseline) are **new** — likely regressions. Flag these prominently.\n4. **Top errors (current period).** Group by `Error Message`, limit to top 10. Include `Error Type`, `File Name`, and unique user count.\n\n#### 2c. User Frustration\n\nUse `Amplitude:query_amplitude_data` to query `[Amplitude] Error Click`:\n\n1. **Error click rate.** Daily error click volume and unique users. Compute current vs baseline.\n2. **Top clicked errors.** Group by `[Amplitude] Element Text` or `[Amplitude] Message`, limit to top 5.\n\n### Phase 3: Page Health Scoring\n\nUse `Amplitude:query_amplitude_data` to score individual pages. Budget: 1-2 calls.\n\n1. Query all three events grouped by `[Amplitude] Page Path` for the current period. For each page, compute:\n   - Network failure count (4xx\u002F5xx `[Amplitude] Network Request` events)\n   - JS error count (`[Amplitude] Error Logged` events)\n   - Error click count (`[Amplitude] Error Click` events)\n   - Unique users affected (across all three)\n\n2. **Score each page.** Assign a health grade:\n\n| Grade | Criteria |\n|-------|----------|\n| **Healthy** | All three signals below product-wide average |\n| **Degraded** | 1-2 signals above average, or any signal >2x average |\n| **Unhealthy** | All three signals above average, or any signal >5x average |\n| **Critical** | Any signal >10x average, or >5% of page visitors affected |\n\n3. Rank pages by severity. Surface the worst 5-10 pages.\n\n### Phase 4: Release Comparison (only if deploy date provided)\n\nIf the user asked about a specific release or if Phase 1 surfaced a deploy that correlates with metric movement:\n\n1. **Before vs after.** Compare the KPIs from Phase 2 using pre-deploy and post-deploy windows instead of the default 7\u002F7 split.\n2. **New errors post-deploy.** Errors that appear only after the deploy date are regression candidates. List them with `Error Message`, `File Name`, and affected user count.\n3. **Endpoints affected.** Compare network failure rates by endpoint pre\u002Fpost deploy.\n4. **Verdict.** Classify the release:\n   - **Clean** — No significant changes in any reliability KPI\n   - **Minor regressions** — 1-2 new errors or small failure rate increases, \u003C1% of users affected\n   - **Significant regressions** — New errors affecting >1% of users, or failure rate increase >50% relative\n   - **Rollback candidate** — Critical new errors, or failure rate increase >100% relative affecting core flows\n\n### Phase 5: Validate\n\nBe the skeptic before presenting:\n\n1. **Partial-day artifacts.** If today is included, compare pace (per-hour rate) not raw totals.\n2. **Day-of-week effects.** Compare same days across weeks. Weekend vs weekday traffic differences can create false signals.\n3. **Bot traffic.** Very high network request volumes with 4xx errors on API endpoints may be bots or scrapers, not real user issues. Note if the pattern looks non-human.\n4. **Expected errors.** 401s on auth endpoints during login flows are normal. 404s on user-generated content URLs are expected. Don't flag these as problems unless they spike.\n5. **Correlation with deployments.** Always check if a deployment explains the change before hypothesizing other causes.\n\n### Phase 6: Build the Report\n\n**Required sections:**\n\n#### 1. Health Summary\n\n```\n## Reliability Report: [Project Name]\nDate: [Today] | Window: [Start] – [End] | Project: [Name] ([ID])\n\n| KPI | Current (7d) | Baseline (7d) | Change | Status |\n|-----|-------------|---------------|--------|--------|\n| Network failure rate | X.X% | X.X% | +X.X% | 🟢\u002F🟡\u002F🔴 |\n| Slow request rate (>3s) | X.X% | X.X% | +X.X% | 🟢\u002F🟡\u002F🔴 |\n| JS error rate (errors\u002Fsession) | X.XX | X.XX | +X.XX | 🟢\u002F🟡\u002F🔴 |\n| Error-free session rate | XX.X% | XX.X% | -X.X% | 🟢\u002F🟡\u002F🔴 |\n| Error click rate (clicks\u002F1K sessions) | X.X | X.X | +X.X | 🟢\u002F🟡\u002F🔴 |\n| Users affected by errors | X,XXX | X,XXX | +XX% | 🟢\u002F🟡\u002F🔴 |\n\n**Overall: [Healthy \u002F Degraded \u002F Unhealthy \u002F Critical]**\n[1-2 sentence summary of the overall state.]\n```\n\nStatus thresholds:\n- 🟢 Stable or improving (change \u003C10% relative)\n- 🟡 Degraded (change 10-50% relative)\n- 🔴 Critical (change >50% relative, or absolute value exceeds acceptable threshold)\n\n#### 2. Release Comparison (conditional)\n\nOnly if the user asked about a release or if a deploy clearly correlates. Use the format from Phase 4 with the verdict.\n\n#### 3. Page Health\n\n```\n### Page Health\n\n| Page | Network Failures | JS Errors | Error Clicks | Users Affected | Grade |\n|------|-----------------|-----------|--------------|----------------|-------|\n| \u002Fpath | XXX | XXX | XXX | X,XXX | 🔴 Critical |\n| \u002Fpath | XXX | XXX | XXX | XXX | 🟡 Degraded |\n```\n\nOnly show degraded, unhealthy, or critical pages. If all pages are healthy, say so in one line.\n\n#### 4. What Changed\n\nNarrative paragraphs (3-5 max) covering the most important changes. Each paragraph follows the pattern:\n\n**[Headline — ≤10 words]** — What changed (with numbers and time context). Why it changed (deployment, experiment, external factor — or \"no clear cause yet\"). Who's affected (segment, page, user count). What to do about it (specific action).\n\nPrioritize:\n1. **New errors** (regressions) over chronic errors (tech debt)\n2. **Worsening trends** over stable-but-bad metrics\n3. **User-facing impact** (error clicks) over silent errors\n4. **Core flow errors** over peripheral page errors\n\n#### 5. Recommended Actions\n\n2-4 numbered items, ordered by urgency. Concrete and copy-paste-ready. Start each with a verb.\n\nFor each action, note the expected impact: \"Fixing the TypeError in `ChartRenderer.tsx` would eliminate ~40% of current JS errors, improving error-free session rate by an estimated 2 percentage points.\"\n\n#### 6. Follow-on Prompt\n\nAsk what to dig into: \"Want me to investigate the chart builder errors in detail, build a reliability dashboard to track these KPIs daily, or compare error rates across plan tiers?\"\n\n---\n\n**Writing standards:**\n\n- **Numbers first.** This is an engineering-oriented report. Lead with data, not narrative. Tables over prose where they're clearer.\n- **Approximate.** \"~2.3%\" not \"2.2847%\". \"~1,200 users\" not \"1,237 users.\"\n- **Always state the time anchor.** \"Over the last 7 days\" or \"since the March 15 deploy.\" Never \"recently.\"\n- **Relative and absolute.** Always provide both: \"Network failure rate increased from 1.2% to 1.8% (+50% relative).\"\n- **Don't bury the lede.** If a release broke something, say it in the first sentence — don't make the reader wade through KPI tables to find it.\n- **Link charts and replays inline** using markdown where available.\n- **Total report length: 500-800 words** for the core report. The KPI table and page health table don't count toward this.\n\n---\n\n## Edge Cases\n\n- **No auto-captured events.** The project may not have Session Replay or autocapture enabled. Report this clearly and suggest enabling it. Describe what data they'd get.\n- **Very low traffic.** If \u003C1,000 network requests in the window, note that sample sizes are too small for reliable rates. Report absolute counts instead of percentages.\n- **All metrics are healthy.** This is a positive finding. Say so clearly: \"All reliability KPIs are stable or improving. Error-free session rate is XX%, network failure rate is X.X%, and no new JS errors have appeared.\" Then surface the top chronic errors as tech debt candidates and suggest an error budget target.\n- **No baseline available.** If the project just started tracking these events, you only have the current period. Skip trend comparisons and present the current snapshot. Recommend checking back in 2 weeks for a meaningful comparison.\n- **User asks \"did the release break anything\" without a date.** Check `use_amp_flags` with `action: \"list_deployments\"` and use the most recent deploy's date. If no deployments are found, ask the user for the deploy date.\n- **Overwhelming number of errors.** If >50K JS errors in the window, focus exclusively on unique error messages and affected user counts. Group by `File Name` first to identify which codebases are noisiest, then drill into error messages within the worst file.\n\n## Examples\n\n### Example 1: Routine Health Check\n\nUser says: \"Give me a reliability check\"\n\nActions:\n1. Get context and project\n2. Query all three error events over 14 days — 7-day baseline vs 7-day current\n3. Compute all KPIs and page health scores\n4. Check deployments for correlation with any regressions\n5. Present the health summary table, worst pages, and top actions\n\n### Example 2: Release Validation\n\nUser says: \"We shipped v4.3 yesterday — did it break anything?\"\n\nActions:\n1. Get context and deployments — find the v4.3 deploy timestamp\n2. Query all three events with pre-deploy (7d before) and post-deploy windows\n3. Identify new errors that didn't exist before the deploy\n4. Compare failure rates and page health pre\u002Fpost\n5. Deliver a release verdict: clean, minor regressions, significant regressions, or rollback candidate\n\n### Example 3: Targeted Area Check\n\nUser says: \"How's reliability on the experiment pages?\"\n\nActions:\n1. Get context\n2. Filter all three events to `[Amplitude] Page Path` containing experiment-related paths\n3. Compute KPIs scoped to those pages\n4. Compare to product-wide averages — is this area better or worse than normal?\n5. Present a focused page health report with actions specific to that area\n",{"name":5,"description":7},"main","9836d94e88404aabb682086405c2aa9cba7657ac","plugins\u002Famplitude\u002Fskills\u002Fmonitor-reliability",156,"https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace","https:\u002F\u002Fgithub.com\u002FCognitionAI","https:\u002F\u002Fcognition.ai\u002F","Makers of Devin, the AI software engineer.","2026-11-13 07:08:23",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add CognitionAI\u002Fdevin-marketplace --skill monitor-reliability",{"tool":39,"command":40},"bunx","bunx skills add CognitionAI\u002Fdevin-marketplace --skill monitor-reliability",{"tool":42,"command":43},"pnpm","pnpm dlx skills add CognitionAI\u002Fdevin-marketplace --skill monitor-reliability","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Famplitude\u002Fskills\u002Fmonitor-reliability\u002FSKILL.md",true,[],[48,54,61,67,74,81],{"skillNo":49,"name":50,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M7BYC4AJGM98P2G9JH7QS24Z","business-data","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fwindsor-ai\u002Fskills\u002Fbusiness-data\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M7BYBZ0CD4CBCPZHRZ551BNK","workflow","Creates durable, resumable workflows using Vercel's Workflow SDK. Use when building workflows that need to survive restarts, pause for external events, retry on failure, or coordinate multi-step operations over time. Triggers on mentions of \"workflow\", \"durable functions\", \"resumable\", \"workflow sdk\", \"queue\", \"event\", \"push\", \"subscribe\", or step-based orchestration.","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fvercel\u002Fskills\u002Fworkflow\u002Fupstream\u002FSKILL.md",[],[],{"skillNo":62,"name":56,"nameZh":6,"description":63,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":64,"categories":65,"keywords":66,"publishedAt":19},"01M7BYB1XSZHKHD8JZTQJ62JN0","Vercel Workflow SDK expert guidance. Use when building durable workflows, long-running tasks, API routes or agents that need pause\u002Fresume, retries, step-based execution, or crash-safe orchestration with Vercel Workflow.","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fvercel\u002Fskills\u002Fworkflow\u002FSKILL.md",[],[],{"skillNo":68,"name":69,"nameZh":6,"description":70,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":71,"categories":72,"keywords":73,"publishedAt":19},"01M7BYASGMK935J33E21N5WTW0","verification","Full-story verification — infers what the user is building, then verifies the complete flow end-to-end: browser → API → data → response. Triggers on dev server start and 'why isn't this working' signals.","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fvercel\u002Fskills\u002Fverification\u002FSKILL.md",[],[],{"skillNo":75,"name":76,"nameZh":6,"description":77,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":78,"categories":79,"keywords":80,"publishedAt":19},"01M7BY9WJERKSKEH9HK6E71GW2","vercel-storage","Vercel storage expert guidance — Blob, Global Config (formerly Edge Config), and Marketplace storage (Neon Postgres, Upstash Redis). Use when choosing, configuring, or using data storage with Vercel applications.","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fvercel\u002Fskills\u002Fvercel-storage\u002FSKILL.md",[],[],{"skillNo":82,"name":83,"nameZh":6,"description":84,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":85,"categories":86,"keywords":87,"publishedAt":19},"01M7BY9N034894YV2J9DCRZHYS","vercel-services","Configure and troubleshoot Vercel Services for multiple frontends and backends in one project. Use when composing a polyglot or multi-service application on one Vercel deployment; defining the `services` key, service-targeted rewrites, or service bindings in `vercel.json`; or running all services with `vercel dev`.","https:\u002F\u002Fgithub.com\u002FCognitionAI\u002Fdevin-marketplace\u002Fblob\u002Fmain\u002Fplugins\u002Fvercel\u002Fskills\u002Fvercel-services\u002FSKILL.md",[],[]]