Replaced the old TEA brief with an indexed knowledge system: the agent now loads topic-specific

docs from knowledge/ via tea-index.csv, workflows reference those fragments, and risk/level/
  priority guidance lives in the new fragment files

src/modules/bmm/testarch/README.md

@@ -19,9 +19,7 @@ last-redoc-date: 2025-09-30
 2. Confirm `bmad/bmm/config.yaml` defines `project_name`, `output_folder`, `dev_story_location`, and language settings.
 3. Ensure a test test framework setup exists; if not, use `*framework` command to create a test framework setup, prior to development.
 4. Skim supporting references (knowledge under `testarch/`, command workflows under `workflows/testarch/`).
-   - `tea-knowledge.md`
-   - `test-levels-framework.md`
-   - `test-priorities-matrix.md`
+   - `tea-index.csv` + `knowledge/*.md`
 
 ## High-Level Cheat Sheets
 
@@ -140,22 +138,25 @@ last-redoc-date: 2025-09-30
 <summary>Command Guidance and Context Loading</summary>
 
 - Each task now carries its own preflight/flow/deliverable guidance inline.
-- `tea-knowledge.md` still stores heuristics; update the brief alongside task edits.
+- `tea-index.csv` maps workflow needs to knowledge fragments; keep tags accurate as you add guidance.
 - Consider future modularization into orchestrated workflows if additional automation is needed.
-- `tea-knowledge.md` encapsulates Murat’s philosophy—update both CSV and knowledge file together to avoid drift.
+- Update the fragment markdown files alongside workflow edits so guidance and outputs stay in sync.
 
 </details>
 
 ## Workflow Placement
 
-We keep every Test Architect workflow under `workflows/testarch/` instead of scattering them across the phase folders. TEA steps show up during planning (`*framework`), implementation (`*atdd`, `*automate`, `*trace`), and release (`*gate`), so a single directory keeps the command catalog and examples coherent while still letting the orchestrator treat each command as a first-class workflow. When phase-specific navigation improves, we can add lightweight entrypoints without losing this central reference.
+The TEA stack has three tightly-linked layers:
+
+1. **Agent spec (`agents/tea.md`)** – declares the persona, critical actions, and the `run-workflow` entries for every TEA command. Critical actions instruct the agent to load `tea-index.csv` and then fetch only the fragments it needs from `knowledge/` before giving guidance.
+2. **Knowledge index (`tea-index.csv`)** – catalogues each fragment with tags and file paths. Workflows call out the IDs they need (e.g., `risk-governance`, `fixture-architecture`) so the agent loads targeted guidance instead of a monolithic brief.
+3. **Workflows (`workflows/testarch/*`)** – contain the task flows and reference `tea-index.csv` in their `<flow>`/`<notes>` sections to request specific fragments. Keeping all workflows in this directory ensures consistent discovery during planning (`*framework`), implementation (`*atdd`, `*automate`, `*trace`), and release (`*nfr-assess`, `*gate`).
+
+This separation lets us expand the knowledge base without touching agent wiring and keeps every command remote-controllable via the standard BMAD workflow runner. As navigation improves, we can add lightweight entrypoints or tags in the index without changing where workflows live.
 
 ## Appendix
 
 - **Supporting Knowledge:**
-  - `tea-knowledge.md` – Murat’s testing philosophy, heuristics, and risk scales.
   - `tea-index.csv` – Catalog of knowledge fragments with tags and file paths under `knowledge/` for task-specific loading.
-  - `test-levels-framework.md` – Decision matrix for unit/integration/E2E selection.
-  - `test-priorities-matrix.md` – Priority (P0–P3) criteria and target coverage percentages.
-  - `knowledge/*.md` – Focused summaries (fixtures, network, CI, etc.) distilled from Murat’s external resources.
+  - `knowledge/*.md` – Focused summaries (fixtures, network, CI, levels, priorities, etc.) distilled from Murat’s external resources.
   - `test-resources-for-ai-flat.txt` – Raw 347 KB archive retained for manual deep dives when a fragment needs source validation.