[Pages] SDM-to-EDM: augmented prompts for plugin + DME remediation

The skill should not modify customer-owned code directly. Plugins live in the
customer's plugin source repo, and Dataverse schema changes should go through
a reviewable solution package — not direct API calls. For both, the script
now generates paste-ready augmented prompts that the user takes to a fresh
Claude Code session, which then does the work and surfaces a diff or artifact
for review before applying.

Plugin remediation prompt:

- Static template at scripts/prompts/plugin-remediation.template.txt with a
  {{PLUGIN_FINDINGS_BLOCK}} placeholder
- Filled at runtime with the customer's actual plugin findings (name, target
  entity, step name) — categorized as Microsoft (no action) / Adxstudio
  (verify V2) / custom (refactor)
- Instructs the receiving session to locate plugin source, refactor entity
  references to powerpagecomponent + inject powerpagecomponenttype filter on
  queries, preserve adx_* attribute references (still valid logical names on
  EDM), update step registration metadata, and show a diff before saving
- Explicit constraints: no production push, no guessing file locations, no
  rewriting GetAttributeValue<T>('adx_name') calls

DME remediation prompt (solution-package approach):

- Static template at scripts/prompts/dme-remediation.template.txt with a
  {{DME_TABLE_GROUPS_BLOCK}} placeholder
- Filled with per-table groupings already produced by
  buildDataModelExtensionChecklists()
- Instructs the receiving session to ask for a publisher prefix, build a
  Dataverse solution package with new custom tables and lookups to
  powerpagecomponent, and produce a .zip ready for pac solution import
- Documents the data-migration step (Power Automate flow OR C# console app
  with batched ExecuteMultipleRequest) — outside solution scope, user runs
  separately
- Explicit constraints: no direct Dataverse API calls, all schema decisions
  reviewed before pack, default to Unmanaged solutions, ask before using
  existing publishers

Surfacing to user (three places):

- Standalone .txt files in <OUTPUT_DIR>/ (plugin-remediation-prompt.txt and
  dme-remediation-prompt.txt) so offline users can access
- Embedded in skill-execution-report.html under collapsible <details> blocks
  with copy-to-clipboard buttons (self-contained navigator.clipboard.writeText
  — no external JS)
- Terminal banner at end of script run with file paths and copy-paste
  instructions

Bug fix in categorizePlugin():

- Old regex (Step name\s*:\s*([^\n]*?)(?:\s\s+Entity Name|$)) mis-handled
  the empty-step case (e.g., Microsoft.Crm.* plugins with no step name),
  greedily capturing 'Entity Name : <entity>' as the step
- Replaced with split-by-2+-whitespace parser + per-field label matching,
  which correctly returns null for empty fields

SKILL.md step 2.2 section 7:

- Replaced the generic 'user must, in the Data workspace...' manual
  remediation with paste-ready augmented-prompt callouts pointing to the
  .txt files and explaining how to use each

DESIGN.md additions:

- New 'Augmented Prompts for Customer-Owned Code' section documenting the
  design rationale (why prompts beat direct execution), template storage
  (scripts/prompts/), runtime substitution, three surfacing locations, and
  coverage table (plugin/DME yes; custom workflows still doc-text-only
  until per-workflow Dataverse queries are added)

Sample reports regenerated against comprehensive-sample.csv (51 findings)
to demonstrate the new augmented-prompt sections end-to-end.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: sparrow1303

plugins/power-pages/skills/migrate-sdm-to-edm/DESIGN.md

@@ -246,12 +246,64 @@ The skill executes local-only analysis (no Dataverse API) for every customizatio
 - **`categorizePlugin(snippet)`** — name-prefix match on `Microsoft.*` / `Adxstudio.*` / custom; emits per-finding action including original entity and step name.
 - **`buildDataModelExtensionChecklists(items)`** — groups column findings by source `adx_*` table; produces one checklist per source table with suggested new-table name and step-by-step guidance from the migration doc.
 
-Output is rendered in `skill-execution-report.html` under the "Liquid Findings — Categorized", "Plugin Findings — Categorized", "Auto-applied Rewrites", and "Data Model Extensions — Per-table Remediation Checklists" sections.
+Output is rendered in `skill-execution-report.html` under the "Liquid Findings — Categorized", "Plugin Findings — Categorized", "Auto-applied Rewrites", "Data Model Extensions — Per-table Remediation Checklists", and the augmented-prompts sections.
 
 See `assets/skill-execution-report.html` for the rendered remediation guidance shown to users.
 
 ---
 
+## Augmented Prompts for Customer-Owned Code
+
+Two customization categories — **custom plugins** and **Data Model Extensions** — involve modifying code or schema that the skill does NOT own:
+
+- **Plugins** live in the customer's plugin source repo (often a separate code repository)
+- **DME (custom columns on adx_* tables)** require Dataverse schema changes that should land via a reviewable solution package, not direct API calls
+
+For both, the skill follows a **paste-ready augmented-prompt** pattern:
+
+1. The script generates a complete, self-contained prompt tailored to the user's actual findings
+2. The prompt is written to a `.txt` file in `<OUTPUT_DIR>/` and embedded in `skill-execution-report.html`
+3. The user opens a fresh Claude Code session pointed at the relevant working directory (their plugin repo for plugins; any working dir for DME)
+4. The user pastes the prompt as the first message
+5. The receiving session performs the work — refactoring plugin code OR building a Dataverse solution package — and surfaces a diff or artifact for the user to review before applying
+
+### Why this design
+
+| Concern | Direct execution from this skill | Augmented prompt approach |
+| --- | --- | --- |
+| Customer-owned plugin source | Skill would need access to the plugin repo — not available | User runs the prompt where the repo is — clean separation |
+| Dataverse schema changes | Direct API calls are hard to undo, hard to review, bypass ALM | Solution package is a reviewable artifact; user imports it themselves |
+| Decision-making (publisher prefix, column types, on-delete behavior) | Lots of interactive prompts in our skill | Batched in the receiving session |
+| Source control | Changes hit Dataverse / customer repo invisibly | All artifacts version-controllable as files |
+
+### Template storage
+
+Prompt templates live as static text files under `scripts/prompts/`:
+
+- `plugin-remediation.template.txt` — placeholder: `{{PLUGIN_FINDINGS_BLOCK}}`
+- `dme-remediation.template.txt` — placeholder: `{{DME_TABLE_GROUPS_BLOCK}}`
+
+The script's `loadPromptTemplate()` reads the file and substitutes the placeholder with the actual findings (markdown-formatted tables / groupings) before writing to `<OUTPUT_DIR>/plugin-remediation-prompt.txt` and `<OUTPUT_DIR>/dme-remediation-prompt.txt`.
+
+### Surfacing to the user
+
+All three locations cover different user contexts:
+
+1. **Terminal output** at the end of script run — visual separator banner with file paths and copy-paste instructions
+2. **Standalone `.txt` files** in `<OUTPUT_DIR>/` — for users who want the prompts without keeping the HTML open
+3. **HTML execution report** — embedded inside collapsible `<details>` blocks with copy-to-clipboard buttons (uses `navigator.clipboard.writeText` — self-contained, no external JS)
+
+### What's covered today
+
+| Category | Prompt? | Notes |
+| --- | --- | --- |
+| Custom plugins | ✅ Yes | Refactor pattern, build/test/deploy guidance, no production push |
+| Data Model Extensions | ✅ Yes | Solution-package output; data-migration step documented but not packaged |
+| Custom-to-adx relationships | ✅ (within DME prompt) | Receiving session can add relationships to the same solution |
+| Custom workflows | ❌ No (yet) | Would require per-workflow Dataverse queries to be useful; generic doc guidance in HTML report for now |
+
+---
+
 ## Component Type Reference Table
 
 For FetchXML and Liquid rewrites, map `adx_*` entity → `powerpagecomponenttype` value:

plugins/power-pages/skills/migrate-sdm-to-edm/SKILL.md

@@ -604,19 +604,29 @@ This pushes the rewritten source files back to the SDM site's Dataverse records,
 
 ### 7. Manual remediation reminders (for non-automatable categories)
 
-Show user the following before proceeding to step 3.1:
+Two customization categories are NOT modified directly by this skill — instead the script generates **paste-ready augmented prompts** that the user takes to a fresh Claude Code session. This is intentional: customer-owned plugin source and Dataverse schema changes should be reviewed and approved before execution.
 
-**Data Model Extensions** — for each custom column found on `adx_*` tables, the user must, in the Data workspace:
+**Plugins on `adx_*` tables** — augmented prompt at `<OUTPUT_DIR>/plugin-remediation-prompt.txt` (and embedded in `skill-execution-report.html`). The prompt:
 
-1. Create a new custom table (e.g., `contoso_webpage`)
-2. Add the custom column (e.g., `contoso_pagetype`) to the new table
-3. Add a lookup column on the new table pointing to `powerpagecomponent`
-4. Migrate data from the old column to the new table
-5. Update any Liquid/FetchXML to reference the new table
+- Includes the verbatim plugin findings from the report (name, target entity, step name)
+- Categorizes each as Microsoft / Adxstudio / custom and includes the refactor pattern only for custom plugins
+- Instructs the receiving Claude session to locate plugin source, refactor to target `powerpagecomponent`, generate a diff for review, get user approval, and only then build
+- Explicit guardrails: no production push, no guessing at file locations, no rewriting `adx_*` attribute references
 
-**Custom-to-adx relationships** — create a new relationship between the custom table and `powerpagecomponent` (e.g., `powerpagecomponent_contoso_pagelogs`).
+To use: tell the user to open a new Claude Code session in their plugin source repo (`claude` command) and paste the contents of the prompt file as the first message.
 
-**Plugins/Workflows on `adx_*` tables** — refactor code to target `powerpagecomponent`, update attribute references, re-register on the new table.
+**Data Model Extensions** — augmented prompt at `<OUTPUT_DIR>/dme-remediation-prompt.txt`. The prompt:
+
+- Includes the per-table groupings from the customization report
+- Instructs the receiving Claude session to ask for a publisher prefix, build a Dataverse **solution package** containing new custom tables, columns, and lookups to `powerpagecomponent`
+- Documents the import + data-migration steps the user runs separately
+- Explicit guardrails: NO direct Dataverse API calls, user must approve schema before pack
+
+To use: tell the user to open a new Claude Code session in any working directory (`claude` command) and paste the contents of the prompt file as the first message. The session produces a reviewable `.zip` for `pac solution import`.
+
+**Custom-to-adx relationships** — currently bundled into the DME prompt. The receiving session can be asked to add the relationship to the same solution package.
+
+**Custom workflows** — no augmented prompt today (would need Dataverse queries to fetch per-workflow primary-entity info). For now, generic guidance shown in the execution report: refactor the workflow to target `powerpagecomponent` and re-register.
 
 Reference: <https://learn.microsoft.com/en-us/power-pages/admin/migrate-enhanced-data-model#considerations-for-site-customization-when-migrating-sites-from-standard-to-enhanced-data-model>