chore(skills): refresh from context-hub (HTML scrapes → clean markdown)#430
chore(skills): refresh from context-hub (HTML scrapes → clean markdown)#430
Conversation
Refreshes wizard's bundled `skills/` from context-hub origin/main + amplitude/context-hub#59 (the HTML→markdown conversion in the build pipeline). The skill payloads sent to the agent on every run drop substantially because reference files are no longer raw HTML scrapes of amplitude.com/docs. ## Why Per-skill references like `references/browser-sdk-2.md` were full HTML pages including `<head>` meta tags, cookie consent JavaScript, top/side nav, inline SVG logos with raw path data, and the "Copy as Markdown" widget. ~80% of those files were page chrome the agent had to swim through before reaching the SDK reference content (init signatures, options, examples). First-turn latency in the wizard was paying for that on every skill load. ## Impact (measured on integration-nextjs-app-router) | file | before | after | reduction | |----------------------------|--------|-------|-----------| | `amplitude-quickstart.md` | 1844 | 209 | 8.8x | | `browser-sdk-2.md` | 4726 | 1834 | 2.6x | | `browser-unified-sdk.md` | 2211 | 322 | 6.9x | Across all 32 integration skills + 5 instrumentation skills + 2 taxonomy skills, the refresh is a net **241K lines removed / 45K added**. No skill directories added or removed — just regenerated content. Critical SDK content (init signatures, fenced code blocks, option tables) verified intact in spot-checks. ## Source context-hub commit base: `435c0f2` (origin/main, 2026-04-27) + the HTML→markdown converter from amplitude/context-hub#59. Once #59 merges and a release ships, this state is reproducible via `pnpm skills:refresh`. ## Test plan - [x] `pnpm test` — 2466 tests pass / 14 skipped (no regressions) - [x] `pnpm lint` clean - [ ] Manual: re-run wizard against the nextjs-app-router test app and verify perceived first-turn latency improves; verify the agent still produces correct SDK init code and event tracking Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
🧙 Wizard CIRun the Wizard CI and test your changes against wizard-workbench example apps by replying with a GitHub comment using one of the following commands: Test all apps:
Test all apps in a directory:
Test an individual app:
Show more apps
Results will be posted here when complete. |
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Broken relative path to taxonomy SKILL.md
- Restored the correct relative path
../../taxonomy/amplitude-quickstart-taxonomy-agent/SKILL.mdin all three instrumentation skill files, replacing the broken../taxonomy/SKILL.mdthat resolved to a non-existent location.
- Restored the correct relative path
Or push these changes by commenting:
@cursor push a8ab29aad3
Preview (a8ab29aad3)
diff --git a/skills/instrumentation/discover-analytics-patterns/SKILL.md b/skills/instrumentation/discover-analytics-patterns/SKILL.md
--- a/skills/instrumentation/discover-analytics-patterns/SKILL.md
+++ b/skills/instrumentation/discover-analytics-patterns/SKILL.md
@@ -25,7 +25,7 @@
When determining naming conventions in this skill, use the following sources in strict order of preference:
1. Events and properties observed from the Amplitude MCP server
2. Real tracking call sites in the codebase
-3. The `taxonomy` skill at `../taxonomy/SKILL.md`
+3. The `taxonomy` skill at `../../taxonomy/amplitude-quickstart-taxonomy-agent/SKILL.md`
---
diff --git a/skills/instrumentation/discover-event-surfaces/SKILL.md b/skills/instrumentation/discover-event-surfaces/SKILL.md
--- a/skills/instrumentation/discover-event-surfaces/SKILL.md
+++ b/skills/instrumentation/discover-event-surfaces/SKILL.md
@@ -23,7 +23,7 @@
that mirror implementation details. Aim for **breadth and quality** — a
downstream skill will narrow the list.
-Read the `taxonomy` skill at `../taxonomy/SKILL.md` to understand core
+Read the `taxonomy` skill at `../../taxonomy/amplitude-quickstart-taxonomy-agent/SKILL.md` to understand core
analytics philosophy and naming standards.
---
diff --git a/skills/instrumentation/instrument-events/SKILL.md b/skills/instrumentation/instrument-events/SKILL.md
--- a/skills/instrumentation/instrument-events/SKILL.md
+++ b/skills/instrumentation/instrument-events/SKILL.md
@@ -24,7 +24,7 @@
with existing patterns, minimal footprint, and properties that actually power
dashboards — not vanity fields nobody queries.
-Read the `taxonomy` skill at `../taxonomy/SKILL.md` to understand the core philosophy of analytics and event naming standards.
+Read the `taxonomy` skill at `../../taxonomy/amplitude-quickstart-taxonomy-agent/SKILL.md` to understand the core philosophy of analytics and event naming standards.
---You can send follow-ups to the cloud agent here.
) The IntroScreen's "Change directory" picker freezes after the user submits a new path. `store.changeInstallDir(newDir)` resets `detectionComplete = false` (so the spinner reappears) and then fires the registered redetector callback — but in production the redetector was never registered, so the spinner spun forever. Same gap meant `store.autoEnableInlineAddons('auto-tui')` never fired in TUI mode. It's only called inside `runFrameworkDetection`, the helper extracted from bin.ts for re-runnability — and that helper had no production caller. `default.ts`'s detection task ran inline as a one-shot IIFE that captured `installDir` from closure and never offered cancellation or re-run. This commit replaces the inline version with the shared `runFrameworkDetection` helper and wires up both: - `setFrameworkRedetector(...)` — closes the IntroScreen loop so submitting a new path actually re-runs detection against it. - `registerActiveDetection(controller)` — lets a directory swap that lands mid-detection cancel the in-flight run instead of leaving a stale `setDetectionComplete()` to fire after the state reset. Removes ~83 lines of duplicated detection logic from default.ts and drops the `FRAMEWORK_REGISTRY` / `detectAllFrameworks` / `DETECTION_TIMEOUT_MS` dynamic imports that fed it. Existing store tests at `src/ui/tui/__tests__/store.test.ts` already cover the `changeInstallDir` + redetector contract end-to-end (lines 1717-1803). All 2469 tests pass. Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…tion skills Applied via @cursor push command
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 2 total unresolved issues (including 1 from previous review).
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Android reference contains only navigation chrome, no documentation
- Removed 33 lines of site-wide navigation links and breadcrumb chrome that the HTML-to-markdown conversion failed to strip, leaving only the actual SDK catalog content.
Or push these changes by commenting:
@cursor push 8d69f87ff3
Preview (8d69f87ff3)
diff --git a/skills/integration/integration-android/references/android.md b/skills/integration/integration-android/references/android.md
--- a/skills/integration/integration-android/references/android.md
+++ b/skills/integration/integration-android/references/android.md
@@ -1,40 +1,6 @@
-[documentation](/docs)
-
-[Get Started](/docs/get-started)
-
-[Data](/docs/data)
-
-[Analytics](/docs/analytics)
-
-[Amplitude AI](/docs/amplitude-ai)
-
-[Session Replay](/docs/session-replay)
-
-[Guides and Surveys](/docs/guides-and-surveys)
-
-[AI Assistant](/docs/assistant)
-
-[Experiment](/docs/experiment-home)
-
-[Admin](/docs/admin)
-
-[Partners](/docs/partners)
-
-[FAQ](/docs/faq)
-
-[SDKs](/docs/sdks)
-
-/
-
-[Amplitude Analytics SDK Catalog](/docs/sdks/analytics)
-
-/
-
-[Android](/docs/sdks/analytics/android)
-
# Android
-`current` [##   Android-Kotlin SDK](/docs/sdks/analytics/android/android-kotlin-sdk)
+`current` [Android-Kotlin SDK](/docs/sdks/analytics/android/android-kotlin-sdk)
[](https://mvnrepository.com/artifact/com.amplitude/analytics-android)
@@ -42,10 +8,10 @@
- [Releases](https://github.com/amplitude/Amplitude-Kotlin/releases)
- [Ampli](/docs/sdks/analytics/android/ampli-for-android-kotlin-sdk)
-`maintenance` [##   Android SDK](/docs/sdks/analytics/android/android-sdk)
+`maintenance` [Android SDK](/docs/sdks/analytics/android/android-sdk)
[](https://mvnrepository.com/artifact/com.amplitude/android-sdk)
- [GitHub](https://github.com/amplitude/Amplitude-Android)
- [Releases](https://github.com/amplitude/Amplitude-Android/releases)
\ No newline at end of file
-- [Migrate to the Android-Kotlin SDK](/docs/sdks/analytics/android/migrate-to-the-android-kotlin-sdk)
+- [Migrate to the Android-Kotlin SDK](/docs/sdks/analytics/android/migrate-to-the-android-kotlin-sdk)You can send follow-ups to the cloud agent here.
Reviewed by Cursor Bugbot for commit b3e688d. Configure here.
| <div class="w-full pt-5 pl-5"> | ||
| <div class="box-border"> | ||
|
|
||
| [FAQ](/docs/faq) |
There was a problem hiding this comment.
Android reference contains only navigation chrome, no documentation
Medium Severity
The HTML-to-markdown conversion for android.md failed to strip site-wide navigation chrome. Lines 1–23 are top-nav links ([Get Started], [Data], [Analytics], etc.) that provide zero value to the agent. The remaining content (lines 25–52) is just an SDK catalog index with GitHub links — no actual SDK initialization, configuration, or API reference content survives the conversion.
Reviewed by Cursor Bugbot for commit b3e688d. Configure here.
…gration SKILL.md The HTML→markdown regen at PR #430 produced workflow lists like: 1. `basic-integration-1.0-begin.md` - Amplitude Setup - Begin The actual files live at references/basic-integration-1.X-*.md. Without the prefix, the agent can't resolve the path when it tries to load the workflow step. The same skill files already use the references/ prefix in their ## Reference files section — this just aligns the workflow section with that convention. 128 refs fixed across 32 integration SKILL.md files.
…egen Three categories of broken references missed by the prior taxonomy fix: 1. discover-analytics-patterns/SKILL.md had a SECOND occurrence of `../taxonomy/SKILL.md` (line 130) — only the first was patched in fd36663. Updates it to `../../taxonomy/amplitude-quickstart-taxonomy-agent/SKILL.md` matching the canonical layout. 2. EXAMPLE.md cross-skill links pointed at non-existent siblings: - integration-javascript_web → `../javascript-node` - integration-astro-static → `../astro-ssr` - integration-nuxt-4 → `../nuxt-3.6` The actual sibling skill dirs are `integration-javascript_node`, `integration-astro-ssr`, and `integration-nuxt-3.6` — and they live one more level up. Updates each link to `../../integration-<name>/`.



Summary
Refreshes wizard's bundled
skills/from context-hub origin/main + amplitude/context-hub#59 (the HTML→markdown conversion in the build pipeline). Skill payloads sent to the agent on every run drop substantially — reference files are no longer raw HTML scrapes of amplitude.com/docs.Why this exists
Per-skill references like
references/browser-sdk-2.mdwere full HTML pages:<head>meta tags, cookie consent JavaScript, top/side nav, inline SVG logos with raw path data, the "Copy as Markdown" widget. ~80% of those files was page chrome the agent had to swim through before reaching the SDK reference content (init signatures, options, examples). First-turn latency in the wizard was paying for that on every skill load.The HTML stripping happens at the source in context-hub#59 (cheerio + turndown in the build pipeline). This wizard PR consumes the resulting cleaner skills.
Impact (measured on
integration-nextjs-app-router)amplitude-quickstart.mdbrowser-sdk-2.mdbrowser-unified-sdk.mdAcross all 32 integration skills + 5 instrumentation skills + 2 taxonomy skills, the refresh is 241K lines removed / 45K added. No skill directories added or removed — just regenerated content. Critical SDK content (init signatures, fenced code blocks, option tables) verified intact in spot-checks.
Source
context-hub commit base:
435c0f2(origin/main, 2026-04-27) + the HTML→markdown converter from amplitude/context-hub#59. Once #59 merges and a release ships, this state will be reproducible viapnpm skills:refresh. Until then, this is an out-of-band refresh — flagging that explicitly in the commit message.Test plan
pnpm test— 2466 tests pass / 14 skipped (no regressions from skill content changes)pnpm lintcleanRelated
cc @amplitude/growth
🤖 Generated with Claude Code
Note
Low Risk
Primarily documentation/reference regeneration and link/path adjustments; low runtime risk, with the main risk being broken references or missing doc content if the conversion output is incomplete.
Overview
Refreshes bundled
skills/content to use cleaned, markdown-only documentation instead of rawamplitude.com/docsHTML scrapes, substantially shrinking reference payloads (e.g.,integration-androiddocs likeamplitude-quickstart.mdandandroid.md).Updates several integration skill
SKILL.mdfiles to point workflow steps atreferences/*paths (instead of root-relative filenames), and adjusts instrumentation best-practices guidance to explicitly treat Browser SDK Autocapture as a fixed excluded event set (with a canonical source to keep in sync).Reviewed by Cursor Bugbot for commit da82fb5. Bugbot is set up for automated code reviews on this repo. Configure here.