Tekion ServiceNow KB — Search & Scrape

The Tekion Knowledge Base lives on ServiceNow at tekion.service-now.com/sp/en.
It is a SEPARATE app from the DMS, BUT it auto-SSOs from Joe's logged-in DMS
session — so via the :9223 persistent (authenticated) browser you can read any
KB article as Joe. (Older memory said the KB was "blocked / PDF-export only" — that
is OBSOLETE. SSO works.)

One-time SSO bootstrap (only if KB shows a login form)

If tekion.service-now.com/sp/en?id=index shows "Login / Log in" instead of
"JC Joe Castelino", establish SSO FIRST. THE RELIABLE METHOD (verified 2026-07-12):
just /navigate to https://app.tekioncloud.com/core/knowledge-base/search —
that's the URL the DMS "Knowledge Base" link fires (discovered via window.open
hook); it performs the full SSO handshake in the SAME tab and lands you on
tekion.service-now.com/sp authenticated as Joe. Requires a live DMS session on
:9223 (verify "Welcome back, Joe" on /home first).

DO NOT waste time on the ServiceNow login page itself — all of these FAIL:

• "Use external login" → User ID + Submit → bounces back to username/password
• login_locate_sso.do with either of Joe's emails → bounces back to login
• The manual Get Help → Knowledge Base click path also has TWO traps: (1) Pendo
  tour overlays swallow the clicks — remove [id*=pendo],[class*=pendo] after
  opening the drawer AND before re-scanning for the "Knowledge Base" element
  (its coords shift once overlays are gone); (2) the link calls window.open
  (new tab the :9223 server doesn't track) — hook
  window.open=u=>{window.__opened.push(String(u))} to capture the URL, then
  /navigate to it in the tracked tab.

The helper script (canonical)

/home/itadmin/tekion-reports/kb_search_scrape.py (uses the :9223 browser):

python3 kb_search_scrape.py search "bin location"          # -> [{kb:KB0012192, title:...}, ...]
python3 kb_search_scrape.py article KB0012192              # -> {url, body}
python3 kb_search_scrape.py article KB0012192 KB0022520    # batch

Tested working 2026-06-29.

Bulk / recursive scraping of a landing-page tree (e.g. "Landing page for
Service 3.0 KB articles", sys_id b2a882cbc3d54350a5d19ec905013167 → ~60 articles):
/home/itadmin/tekion-kb/scripts/scrape_service30_tree.py — BFS from a seed
sys_id, extracts 32-hex sys_ids from body links (entityId=...), follows
"Landing page for ..." sub-indexes to depth 2, saves each to
tekion-kb/text/SVC30_<sysid8>_<slug>.txt. Adapt SEED for other trees. Run as
background (8s page-load + 2s pacing per article). It self-detects auth loss
("Log in" in document.title → prints AUTH/EMPTY, doesn't save garbage).

How it works (the mechanics — the hard-won parts)

1. The KB is a Seismic web-component SPA with ~390 shadow roots. The search box
   is an <input> inside sn-search-combobox → must PIERCE shadow DOM to find it.
   Normal document.querySelector('input') returns nothing.
2. Find the search input: recurse shadow roots collecting visible INPUTs; on
   the home page it sits around (617,287). Click it with /mouse (raw coords) to focus.
3. Type into it: document.activeElement (drill through .shadowRoot.activeElement)
   → set value via the native HTMLInputElement.prototype.value setter, dispatching
   input events per char, then a synthetic Enter KeyboardEvent. This navigates to
   ?id=search&q=.... (The :9223 server has NO keyboard endpoint — must do it in-page.)
4. Read results: deep-walk shadow roots concatenating textContent; results render
   as triples KBxxxxxxx / "N ago" / Title. Pair each KB\d{7} with the next title line.
   Left rail also shows KB Category facets (Parts & Inventory, Warehouse Management,
   Parts Settings, Physical Inventory, Repair Order, etc.).
5. Open an article by KB number (THE reliable trick): navigate directly to
   https://tekion.service-now.com/sp/en?id=kb_article_view&sysparm_article=KBxxxxxxx
   — it 302s to the article slug. DO NOT try clicking result links: they're
   javascript:void(0) Angular handlers and .click() won't route.
   sys_id articles too: many articles/landing pages are addressed by 32-hex
   sys_id instead of KB number — ?id=kb_article&sys_id=<32hex> works the same.
   Links INSIDE article bodies use the form
   https://app.tekioncloud.com/core/knowledge-base?entityType=KNOWLEDGE&entityId=<32hex>
   — the entityId IS the sys_id; extract it and rewrite to the sp/en URL.
6. Read article body from REGULAR DOM (not shadow): .kb-article-content
   (fallbacks [itemprop=articleBody], article, .article-content). Body is plain text.

Storing into the brain (the ingest workflow)

After scraping, persist so it's searchable in GBrain:

1. Save raw text → /home/itadmin/tekion-kb/text/<KB>_<slug>.txt
2. Write a DISTILLED markdown page → /home/itadmin/tekion-kb/distilled/<topic>.md
   (steps, settings, exact field names, gotchas, + a KB# index table).
3. Auto-embed: anything dropped in /home/itadmin/tekion-kb/{distilled,text,transcripts}
   is picked up by session-end-sync.sh → ingest-kb-batch-to-brain.py within ~15 min
   (sha1-tracked, only new/changed). No manual embed step. Search later via gbrain.
4. If it documents a workflow, also create/patch a tekion-* skill + update tekion-sitemap.

Pitfalls

• Full page navigation WIPES any injected XHR/fetch hooks — reinstall after load, or
  just use the direct article URL method (no hooks needed).
• The ServiceNow Table API (/api/now/table/kb_knowledge) returns 401; the SP search
  POST (/api/now/sp/search) needs a portal+source payload that's fiddly — the UI
  search-box-typing method above is more reliable than reverse-engineering the API.
• Verify store/auth isn't the blocker: if KB shows login, do the Get Help bootstrap.
• Use vision (browser_vision/:9223 /screenshot + vision_analyze) to read result
  positions and KB numbers when DOM parsing is ambiguous — the rail tabs are clickable
  category filters worth exploring (Joe's tip 2026-06-29).

Service 3.0 corpus (scraped 2026-07-13 — already in the brain)

The full Service 3.0 KB tree (seed sys_id b2a882cbc3d54350a5d19ec905013167) is
scraped: 76 articles at /home/itadmin/tekion-kb/text/SVC30_*.txt, embedded in
GBrain — search there before re-scraping. Derived deliverables (sent to Joe as a
Gmail DRAFT with 6 PDFs, 2026-07-13):

• /home/itadmin/tekion-kb/distilled/service-3-0-kb-index.md — index of all 76
• /home/itadmin/tekion-kb/distilled/service-3-0-reading-action-plan.md — Joe's 4-session (~5.5h) dependency-ordered reading path
• /home/itadmin/tekion-kb/distilled/service-3-0-lesson-plan.md — role-based training tracks (A Advisors, B Cashiers/Office = most-changed, C Managers = go FIRST/own config, D BDC), 2 wks/store, go-live gates
• /home/itadmin/tekion-kb/cheatsheets/ — 4 print-ready 1-pagers (advisor / cashier / accounting / manager)

CORE 3.0 CONCEPT (the unlock for any Service 3.0 question): payers are split
from pay types — one CP job can carry MULTIPLE payers; each payer gets its own
invoice, its own real-time cashiering line (2.0 incremental cashiering is gone),
and its own JE posting line. Tax codes become MANDATORY. Store setup dependency
order: Pay Types Setup → Tax Codes → Vehicle Groups → Labor Pricing → Opcode
pay-types → Customer Mgmt defaults → Fees. Landing pages (12 of the 79 sys_ids)
are just link hubs — skip when reading.

Cross-references

• tekion-sitemap, persistent-browser-server, tekion-browser-navigation (dealer switch),
  jay-brain-and-skill-index (GBrain ingest internals).