mirror of
https://github.com/forcedotcom/afv-library.git
synced 2026-08-04 22:33:31 +08:00
14c5adeaf2
Modern developer.salesforce.com guide pages now publish a raw-Markdown twin — swap the .html extension for .md on the same URL (the page's "Copy/View as Markdown" links expose it) and a plain web fetch returns clean Markdown with no JS shell. That's faster and lighter than browser-rendered extraction, so the fetching-salesforce-docs playbook should try it first for those pages. Documents the scope: it's confirmed for modern guide pages under /docs/platform/.../guide/*, does NOT work for legacy atlas.* pages (which return an HTML shell), and help.salesforce.com still needs the Playwright extractor — fall back to the existing playbook in those cases. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
75 lines
2.8 KiB
Markdown
75 lines
2.8 KiB
Markdown
# fetching-salesforce-docs
|
|
|
|
## What it is
|
|
|
|
`fetching-salesforce-docs` is a **prompt-only skill**.
|
|
|
|
It gives a practical retrieval playbook for official Salesforce docs on the public web, especially when:
|
|
- `developer.salesforce.com` pages are JS-heavy
|
|
- `help.salesforce.com` pages return shell content
|
|
- `architect.salesforce.com` / `admin.salesforce.com` pages need browser-rendered extraction
|
|
- `lightningdesignsystem.com` pages contain official SLDS guidance
|
|
- the real answer is on a child page, not the guide homepage
|
|
|
|
## What it is not
|
|
|
|
This skill does not include:
|
|
- local corpus workflows
|
|
- indexing
|
|
- benchmark workflows
|
|
- any required helper CLI dependency
|
|
- PDF fallback guidance
|
|
|
|
## Use it for
|
|
|
|
- official Salesforce docs lookup
|
|
- hard-to-fetch Help articles
|
|
- Apex / API / LWC / Agentforce documentation grounding
|
|
- deciding when to follow child links from broad official guide pages
|
|
- rejecting weak results such as shells, landing pages, and third-party summaries
|
|
|
|
## Markdown shortcut (developer.salesforce.com)
|
|
|
|
Modern developer-guide pages publish a raw-Markdown twin: swap `.html` → `.md` on the same URL
|
|
and fetch it with a **plain web fetch** — no Playwright needed. The page's "Copy as Markdown" /
|
|
"View as Markdown" links expose this. Try it **first** for `developer.salesforce.com/docs/platform/.../guide/*`
|
|
pages. It does **not** work for legacy `atlas.*` pages (those return an HTML shell), and
|
|
`help.salesforce.com` still needs the renderer — use the extractor below for those.
|
|
|
|
## Optional utility
|
|
|
|
A tiny wrapper is available for official Salesforce doc URLs:
|
|
|
|
```bash
|
|
python3 skills/fetching-salesforce-docs/scripts/extract_salesforce_doc.py \
|
|
--url "https://help.salesforce.com/s/articleView?id=service.miaw_security.htm&type=5" \
|
|
--pretty
|
|
```
|
|
|
|
Behavior:
|
|
- automatically routes `help.salesforce.com` URLs into the dedicated Help extractor
|
|
- supports official Salesforce-owned doc hosts such as `developer.salesforce.com`, `architect.salesforce.com`, `admin.salesforce.com`, `lightningdesignsystem.com`, and other official Salesforce documentation pages
|
|
- supports optional best-effort stealth mode via `--stealth`
|
|
|
|
Dependencies for the helper scripts live in:
|
|
- `skills/fetching-salesforce-docs/requirements.txt`
|
|
|
|
The installer sets up an isolated runtime under `~/.claude/.fetching-salesforce-docs-runtime`, installs those Python packages there, and installs the Playwright Chromium browser automatically during install/update.
|
|
|
|
The underlying Help extractor is also available directly at:
|
|
|
|
```bash
|
|
python3 skills/fetching-salesforce-docs/scripts/extract_help_salesforce.py \
|
|
--url "https://help.salesforce.com/s/articleView?id=service.miaw_security.htm&type=5" \
|
|
--pretty
|
|
```
|
|
|
|
## Key idea
|
|
|
|
Keep retrieval:
|
|
- **official-source-first**
|
|
- **HTML-only**
|
|
- **targeted**
|
|
- **child-link aware**
|
|
- **strict about exact concept matching**
|