Run any Skill in Manus with one click

sentry-web-vitals

Stars9

Forks0

UpdatedJune 21, 2026 at 06:13

Fetch and compare per-route p75 Web Vitals (LCP/FCP/CLS) from Sentry Discover for the daily-writing-friends production project, with an F2-weighted score and pre/post-deploy diff. Use when the user wants to verify whether performance improved or regressed after a deploy, compare pre/post-merge windows for a PR, identify the worst-performing routes for the next perf push, or sanity-check a Web Vitals claim. Triggers - "web vital 확인해줘", "p75 비교", "회귀 있나", "perf 점수", "이 PR 머지 후 LCP 어떻게 됐어", "sentry에서 가져와".

Installation

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

Run Skill in Manus

Source

BumgeunSong

BumgeunSong/daily-writing-friends

View GitHub Repository View Creator Repositories

Download

Run Skill in Manus

Related occupationsSOC

Based on SOC occupation classification

Software DevelopersComputer and Mathematical Occupations·SOC 15-1252

File Explorer

4 files

SKILL.md

readonly

name

sentry-web-vitals

description

Sentry Web Vitals analysis

What this skill does

Pulls per-route p75 LCP/FCP/CLS from Sentry Discover, normalizes route IDs, and prints a markdown comparison table with deltas and an F2-weighted score. Used to verify performance changes against production data, not synthetic harness output.

Prerequisites

SENTRY_AUTH_TOKEN env var set (Sentry user auth token with org:read + event:read)
python3 + curl on PATH
Org/project default to bumgeun-song / 4508460981747712 — override with SENTRY_ORG / SENTRY_PROJECT_ID if needed

Quick workflow

For a "did this PR actually improve perf?" question, run the comparison script with a pre-merge baseline and a post-deploy window:

scripts/compare.py 2026-05-28T00:00:00 2026-06-04T14:09:06 \
                   2026-06-13T08:06:50 2026-06-21T05:27:00

Output is a markdown table with route, sample size, p75 LCP/FCP/CLS pre→post with percentage delta, and the F2 score per route + a count-weighted overall.

For a one-off snapshot (no comparison), use the raw query:

scripts/query.sh 2026-06-13T00:00:00 2026-06-21T00:00:00 | python3 -m json.tool

Picking time windows

Baseline: pick a 7-day window before the change merged. Use the merge commit time from gh pr view <num> --json mergedAt.
Post: start from the deploy completion time, not merge time. End at "now" or whenever you want to cut.
Minimum length: 4 days post-deploy. Less than that and /board/*/post/* will have <10 LCP samples and any delta is noise.
Window matching is NOT required — the script reports normalized per-route p75, so 7d vs 4d is fine (the comparison is rate-independent).

Interpreting the output

Sample column shows LCP-measured pageloads per route. The marker ⚠️ means n<20 — treat the p75 as directional, not a number to act on.
F2 score uses the same weighting as the local perf harness (LCP 0.5 + FCP 0.2 + CLS 0.3) with Google thresholds (Good=1.0, Needs Improvement=0.5, Poor=0). The count-weighted overall at the bottom is what most closely tracks the harness's "overall F2" claim.
Volume drop post-deploy is normal in this app — many users sit on stale PWA service worker builds for several days. Do NOT chase it. Trust per-route p75 when n≥20.

When the data looks wrong

If the post-window has <10 samples on the routes you care about after >7 days, the cause is usually:

Sample rate was reverted to 0 or unset (check apps/web/src/sentry.ts — SENTRY_CONFIG.TRACE_SAMPLE_RATE should be ≥0.2)
BrowserTracing integration removed entirely (check integrations: includes reactRouterV6BrowserTracingIntegration)
wrapCreateBrowserRouterV6 not used (check router.tsx uses sentryCreateBrowserRouter)

See references/methodology.md for thresholds, route aggregation rules, the boardLoader non-pageload gotcha, and the F2 formula.

More from this repository

same repository

run-web

BumgeunSong/daily-writing-friends

Launch recipe for the daily-writing-friends web app (apps/web — Vite + React + React Router, dev server on http://localhost:5173). Use to build and run the app locally so /run and /verify can drive the real UI. Captures the mandatory nvm node-22 PATH prefix (the default Homebrew node is broken), the dev-server command and port, backend selection (cloud vs local Supabase), and the static gate commands (type-check, tests).

2026-06-289

verify-browser

BumgeunSong/daily-writing-friends

Use after UI/frontend changes — layout, responsive behavior, navigation/routing, or any user-facing flow — to prove it actually works by driving a real headless browser (Playwright) against the running dev server. Asserts behavior (clicks, URL changes, visible text), measures layout (alignment via bounding boxes, overflow via scrollWidth minus clientWidth), captures console/page errors, and saves screenshots. Reach for this whenever "it type-checks and the unit tests pass" does not prove the feature renders and behaves correctly in a browser. Complements verify-runtime (which checks data-flow via dev logs).

2026-06-289

composable-layout

BumgeunSong/daily-writing-friends

Use when writing or modifying page-level layout markup in apps/web — max-width containers, flex rows, vertical spacing — or composing multiple sections on a page. Enforces shared/ui layout primitives (ReadingColumn, Stack, Row) and compound-component slots over scattered Tailwind classes.

2026-06-219

improving-web-vitals

BumgeunSong/daily-writing-friends

Use when the user wants to improve daily-writing-friends production Web Vital scores (LCP/FCP/CLS) by running additional iterations of the local perf harness to drive the overall F2 score higher. Triggers - "web vital 개선", "LCP 줄여", "F2 점수 올려", "harness 한 번 더 돌려", "perf 개선 작업", "perf iteration", "drive score up".

2026-06-219

integration-testing

BumgeunSong/daily-writing-friends

Use when writing tests for React components or custom hooks that touch the React Query cache, MSW-mocked Supabase, or React Router data-router loaders. Triggers - writing component tests, writing hook tests that need a provider tree, creating MSW handlers, render() + screen assertions, data-fetching component tests, form integration tests, loader/errorElement tests, optimistic-mutation tests. Does NOT cover pure-function unit tests (use the testing skill) or cross-page journeys (use Playwright E2E).

2026-06-149

daily-writing-friends-design

BumgeunSong/daily-writing-friends

Design system for Daily Writing Friends app. MUST use when doing ANY UI work including components, pages, buttons, forms, styling, Tailwind CSS, dark mode, theming, layouts, cards, inputs, or visual changes. Ensures consistent design tokens, button hierarchy, and accessibility.

2026-06-149

name

sentry-web-vitals

description

Sentry Web Vitals analysis

What this skill does

Prerequisites

SENTRY_AUTH_TOKEN env var set (Sentry user auth token with org:read + event:read)
python3 + curl on PATH
Org/project default to bumgeun-song / 4508460981747712 — override with SENTRY_ORG / SENTRY_PROJECT_ID if needed

Quick workflow

For a "did this PR actually improve perf?" question, run the comparison script with a pre-merge baseline and a post-deploy window:

scripts/compare.py 2026-05-28T00:00:00 2026-06-04T14:09:06 \
                   2026-06-13T08:06:50 2026-06-21T05:27:00

Output is a markdown table with route, sample size, p75 LCP/FCP/CLS pre→post with percentage delta, and the F2 score per route + a count-weighted overall.

For a one-off snapshot (no comparison), use the raw query:

scripts/query.sh 2026-06-13T00:00:00 2026-06-21T00:00:00 | python3 -m json.tool

Picking time windows

Baseline: pick a 7-day window before the change merged. Use the merge commit time from gh pr view <num> --json mergedAt.
Post: start from the deploy completion time, not merge time. End at "now" or whenever you want to cut.
Minimum length: 4 days post-deploy. Less than that and /board/*/post/* will have <10 LCP samples and any delta is noise.
Window matching is NOT required — the script reports normalized per-route p75, so 7d vs 4d is fine (the comparison is rate-independent).

Interpreting the output

Sample column shows LCP-measured pageloads per route. The marker ⚠️ means n<20 — treat the p75 as directional, not a number to act on.
F2 score uses the same weighting as the local perf harness (LCP 0.5 + FCP 0.2 + CLS 0.3) with Google thresholds (Good=1.0, Needs Improvement=0.5, Poor=0). The count-weighted overall at the bottom is what most closely tracks the harness's "overall F2" claim.
Volume drop post-deploy is normal in this app — many users sit on stale PWA service worker builds for several days. Do NOT chase it. Trust per-route p75 when n≥20.

When the data looks wrong

If the post-window has <10 samples on the routes you care about after >7 days, the cause is usually:

Sample rate was reverted to 0 or unset (check apps/web/src/sentry.ts — SENTRY_CONFIG.TRACE_SAMPLE_RATE should be ≥0.2)
BrowserTracing integration removed entirely (check integrations: includes reactRouterV6BrowserTracingIntegration)
wrapCreateBrowserRouterV6 not used (check router.tsx uses sentryCreateBrowserRouter)

See references/methodology.md for thresholds, route aggregation rules, the boardLoader non-pageload gotcha, and the F2 formula.