onetyler-foundry-team-agent-kb

Knowledge corpora for the OneTyler Cloud Living Foundry Team agent and its five sub-agents. Each Knowledge-<Domain>/ folder is a deployment surface: its files are what the corresponding Foundry agent retrieves from its tenant knowledge-base collection. Adding, removing, or renaming a file changes what that agent knows.

This README is the team-level routing layer. Each folder additionally has its own _START_HERE.md, which is the routing guide within that corpus. The Team agent reads this file to pick a sub-agent; the sub-agent reads its own _START_HERE.md to pick a file.

---

Running the review tool

Reviewing transcripts is a double-click — no assistant needed. Turning that feedback into knowledge-file changes does need one, and that is the actual goal; the Save & Publish page generates the prompt for it. See DAILY-WORKFLOW.md for the whole loop.

macOS	double-click Start-reviewing.command
Windows	double-click Start-reviewing.bat
Any terminal	python3 scripts/start.py (Windows: python scripts\start.py)

It brings the repo up to date, pulls new conversations from Foundry, starts the review UI on http://127.0.0.1:7777 and opens your browser. Leave the window open while you review; Ctrl-C stops it.

Loopback only — the address works on your machine and nowhere else.

• If 7777 is taken, add --port 7778.
• If FOUNDRY_API_KEY is not set it skips the fetch and you review what is already here.
• It never discards your work. With unsaved reviews it leaves the branch alone and says so.

Full walkthrough, including what to do after you send reviews in and the admin-only Foundry steps: DAILY-WORKFLOW.md.

---

New here?

You are	Start with
New to using AI assistants at all	contributor-prompting-guide.md — a plain-English phrasebook: what to say to get what you want. No technical background needed.
A contributor setting up for the first time	contributor-initial-prompt.md — the whole file is the prompt. Select all, copy, paste to your AI. Nothing else in it.
A contributor whose setup needs updating	contributor-update-prompt.md — the whole file is the prompt, same as above. Use when an admin says the process changed.
A reviewer wanting the walkthrough	transcripts/ONBOARDING.md
Looking for the process and field definitions	transcripts/README.md
An AI agent working in this repo	CLAUDE.md — read it in full before anything else

Who may change what. Contributors own subject-matter content; admins own anything that decides which agent answers. The authoritative list is .github/admin-only-paths.txt.

Contributors may change	Admin-only
Knowledge content in any Knowledge-<Domain>/ folder — the Conf-, Docusaurus-, FAQ-, Misc-, Training-, GitHub- files	README.md (this file — the team routing table), team-config/, every Knowledge-*/_START_HERE.md
Review verdicts under transcripts/	CLAUDE.md, contributor-initial-prompt.md, transcripts/README.md, transcripts/ONBOARDING.md, scripts/, templates/, .github/, .gitignore, contributors.json

_START_HERE.md sits in a folder contributors can otherwise edit, but it carries cross-agent hand-off rules, which is team-level routing — so it is admin-only.

Enforced by CODEOWNERS plus branch protection, a CI check, and a warning in start_review_session.sh. Found a problem in an admin-only file? Say so in your PR description rather than fixing it.

What no check catches: routing advice written inside a knowledge file. Human review is the only control for that.

---

Team composition

Foundry team: OneTyler Cloud Living — Amazon Bedrock, Claude 4.5 Sonnet, temperature 0.7.

Sub-agent (Foundry name)	Corpus folder in this repo	Owner
Ops Center	Knowledge-OpsCenter/	this repo
Support Access Center	Knowledge-SupportAccessCenter/	this repo
General Blueprint Docs Agent	Knowledge-BP-General/	this repo
Aligned Releases	Knowledge-AlignedReleases/	this repo
Tyler Identity Assistant	Knowledge-TylerIdentity/	this repo

Tyler Identity was cut over to this repo on 2026-08-24: its collection now holds Docusaurus-Identity.md, _START_HERE.md, Conf-IdentityTickets.md, FAQ-Identity.md and the shared ticket catalog. Docusaurus-Identity.md still reflects a Blueprint pull of 2026-05-20 — re-derive it when Blueprint moves.

In addition, Knowledge-Shared/ holds content every agent needs and is uploaded to all writable collections — currently the OneTyler ticket catalog. See its _START_HERE.md.

---

Routing table — which sub-agent gets the question

Route on the user's intent, not on incidental keyword matches. A question that merely mentions Ops Center while asking about identity federation belongs to Identity.

Route to	When the question is about	Trigger keywords
Ops Center	Org/workspace lifecycle, product licensing & activation, org import/create, CRM customer identifiers, Ops Center permissions & telemetry, environments & allow-listing, TCP webhooks, WM→WD migration	"Ops Center", org key, workspace, licensing, availability, product registration, CRM identifier, allow-list
Support Access Center	Time-bound Tyler-staff access to customer installations, SAC groups, access request/approval/extension/revocation, SAC product integration (Security API + revoked webhook), access history & auditing	"SAC", "Support Access Center", support request, access approval, extend access, Support Council
Tyler Identity Assistant	Identity Workforce/Community, Gateway, Workforce Direct/Managed/Delegated configuration, federation, credential templates, login & token flows	"Identity", "Gateway", Workforce Direct/Managed/Delegated, federation, IdP, OIDC, SSO
Aligned Releases	Release Manager (the internal tool for authoring client-facing features), quarterly GA release model, feature lifecycle (Planned/Private Preview/Public Preview/GA), taking a feature GA, cohorts and cohort assignment, feature activation windows, release documentation and notes, versions like 2026.1, maintenance windows, the Aligned Releases API and SDK, client release notification emails	"release manager", "release management", "aligned release", "cohort", "feature", "feature activation", "feature flag", "release", "GA", "private preview", "public preview", "2026.1", release notes
General Blueprint Docs Agent	Everything else in Tyler Blueprint / TCP: platform orientation & glossary, client & ops applications, TCP/TID API catalog, service architecture, DevOps, platform security, Status Page & SLA	"Blueprint", docs.tylerdev.io, glossary/terminology, Admin Center, architecture, SLA

Routing rules

0. Ticket questions always go to Ops Center. Any "which ticket do I file / how do I request access or permissions" question — including Identity and Support Access Center tickets — is answered from Knowledge-Shared/Conf-OneTylerTickets.md, the single authoritative catalog. One copy, not three, so the copies cannot drift. Never invent a ticket URL.

1. Definitions are not hand-offs. Defining a term (e.g. what "Workforce Managed" means) is answerable from the BP-General glossary. Hand off to a specialist when the user needs workflows, configuration, or deep how-to in that domain.

2. General Blueprint Docs Agent is the default. If no specialist clearly owns the question, it goes here — this corpus is explicitly scoped as "the rest of Blueprint/TCP."

3. Ops Center vs SAC. Both touch customer access. Ops Center = provisioning and lifecycle of orgs/workspaces/products. SAC = time-bound staff access into an already-provisioned customer installation.

4. Aligned Releases vs Ops Center. Aligned Releases refers to productRegistrationId and workspaceKey constantly but does not own them. "What is a product registration" → Ops Center. "Assign this workspace to cohort 3" → Aligned Releases.

   "Release Manager" is ALWAYS Aligned Releases. It is the internal tool where Tyler staff author the features clients see, assign cohorts, and manage the feature lifecycle. The name invites two wrong guesses — that it is an Ops Center screen, or that "release manager" means a person's job title. Neither. Any question naming it, however phrased, goes to Aligned Releases. Same for "release management".

   Note the client-facing half lives in Admin Center, so a question about what a client sees about an upcoming release is still Aligned Releases even though Admin Center is normally a BP-General topic. Release Manager is internal; Admin Center is where clients see the result.

4a. Disambiguating the broad keywords: "feature" and "release". Both are Aligned Releases triggers and both appear constantly in unrelated questions, so use the sense, not the word:

Ask	Route
"What features does Ops Center have?" / "does this product support X?"	not Aligned Releases — product capability, so Ops Center or BP-General
"How do I add a feature / take a feature GA / put it in public preview?"	Aligned Releases
"When is the next release / which cohort activates when?"	Aligned Releases
"How do I release my product to a new customer?" (licensing/activation)	Ops Center

The tell for Aligned Releases is the release-train vocabulary around the word — cohort, activation window, preview stage, GA date, Release Manager, release notes. Bare "feature" or "release" with none of that is usually somebody else's question. 5. Ops Center vs Identity. Ops Center covers what an operator does in the Ops Center UI (including retargeting a workspace's gateway). Identity covers how the identity system itself is configured and how tokens/federation work.

Technical and API questions about organizations, workspaces, licensing or availability go to Ops Center, not Identity — including listing or searching them programmatically, where the answer is the TCP Search API (Docusaurus-OpsCenterAdoption.md → Listing Ops Center Organizations and Workspaces). Mentioning tokens or credentials for the call does not make it an Identity question: Identity owns how you authenticate, Ops Center owns what you call. Recorded after transcript identity/2026-08-27--f4a25651, where Identity answered "how do I get a list of workspaces for an organization using the api?" itself while the correct content sat indexed in the Ops Center corpus. 6. Don't split one question across two agents. Pick the owner of the user's actual goal; the answering agent can name the other domain if a follow-up is needed.

---

Naming conventions

Filename prefixes encode the source system, which signals authority and freshness:

Prefix	Source
Conf-	Confluence (tylertech.atlassian.net/wiki/)
Docusaurus-	Tyler Blueprint (docs.tylerdev.io) — the published source of truth
Training-	Distilled from official training assets (videos, decks, PDFs)
GitHub-	GitHub repo content (github.com/tyler-technologies/...)
Misc-	Curated bookmark catalog spanning multiple source systems
FAQ-	Authored here — no upstream source. One per agent corpus. The home of record for answers that exist nowhere else: SME guidance given verbally, behaviour learned by observation, corrections an upstream owner has not yet made. Everything else in a corpus is re-derived from its source; these entries would be lost by that, so they live apart and carry their own provenance.
_START_HERE.md	Per-corpus routing guide; leading underscore sorts it first

Files are the GPT-optimized form, not raw scrape output: clean markdown, a decision guide near the top, a glossary where relevant, self-contained sections (RAG retrievers chunk independently of headings), and direct URLs preserved verbatim.

---

Deploying to Foundry

Knowledge files are uploaded to the tenant knowledge base as plain text/markdown. There is no update-in-place endpoint — updating content means upload new, delete old, then sync.

Non-obvious constraints that will bite you:

• ingestionStatus: "ingested" proves nothing. Always verify with a scoped retrieval probe (POST /api/tenant-knowledge-base/retrieve with filterCollectionNames). Markdown avoids the silent-failure classes that affect PDFs.
• Uploads cap at 10 files per request and each request auto-triggers a sync. Bedrock runs one ingestion job at a time and only indexes files present when its scan starts. Upload all batches first, then trigger one consolidated sync.
• Re-sync cannot repair an already-indexed file. Delete and re-upload instead.
• Metadata tags support exact-match, AND-combined filters only — no ranges, substrings, or OR.

---

Index hygiene

When you add, rename, or remove a file:

1. Update that folder's _START_HERE.md (file catalog + routing table). A stale start page actively misleads the agent.
2. Update this README if the change affects team-level routing or adds a corpus.
3. Re-upload the changed files and re-verify retrieval.