Plugin check: Run node "${PLUGIN_ROOT}/scripts/check-version.js" — if it outputs a message, show it to the user before proceeding.

Integrate Web API

Integrate a Power Pages site with Dataverse. SPA sites retain the framework-specific client, service, and hook workflow. Non-SPA sites use native Forms, Lists, and Liquid first, adding Web API JavaScript only when the requested client interaction requires it.

Site Type Routing

Before reading or executing the workflow below, perform the existing SPA project lookup first:

• Look for powerpages.config.json in the current directory or immediate subdirectories using the existing file-search patterns.

• If found, continue with the existing SPA workflow below. Do not run any non-SPA site detection or change the SPA phases, prompts, gates, generated files, or deployment behavior.

• Only when that existing SPA lookup finds no powerpages.config.json, run:

  node "${PLUGIN_ROOT}/scripts/lib/detect-project-context.js"
  

  Copy

• If it returns siteType: "non-spa", read and follow the complete seven-phase workflow in ${PLUGIN_ROOT}/skills/integrate-webapi/references/non-spa-site-workflow.md. Do not execute any of the SPA workflow below.

• Otherwise, continue to Phase 1 and preserve its existing not-found handling.

The non-SPA route supports request-focused and discovery invocation as defined in its workflow reference. SPA invocation semantics remain those defined by the existing workflow below. When $ARGUMENTS contains [CALLED-BY-PARENT-SKILL], pass that sentinel and its optional caller=<skill-name> token into the non-SPA workflow. It suppresses commits and deployment handoffs only; it does not bypass the non-SPA scope approval or security planning.

Core Principles

• First table sequential, then parallel: The first table must be processed alone because it creates the shared powerPagesApi.ts client. Once that exists, remaining tables can be processed in parallel since each creates independent files (types, service, hooks).
• Parallelize independent agents: The table-permissions-architect and webapi-settings-architect agents are independent — invoke them in parallel rather than sequentially.
• Permissions require deployment: The .powerpages-site folder must exist before table permissions and site settings can be configured. Integration code can be written without it, but permissions cannot.
• AI-only read mode is opt-in: When invoked by another skill (e.g. /add-ai-webapi) with the [AI-READ-ONLY] sentinel in $ARGUMENTS, the flow produces read-only code and hardens the settings/permissions for AI summarization reads. See Phase 1.6 for the contract. Human invocations never trigger this mode.
• Use TaskCreate/TaskUpdate: Track all progress throughout all phases — create the todo list upfront with all phases before starting any work.
• Preserve site integrity: Before an agent changes mock-backed or data-driven UI, read ${PLUGIN_ROOT}/references/site-modification-integrity.md; localize every new visible state when a manifest exists and keep the component bidirectional and expansion-safe.

Prerequisites:

• An existing Power Pages SPA site created via /create-site
• A Dataverse data model (tables/columns) already set up via /setup-datamodel or created manually
• The site must be deployed at least once (.powerpages-site folder must exist) for permissions setup

Initial request: $ARGUMENTS

Workflow

1. Verify Site Exists — Locate the Power Pages project and verify prerequisites
2. Explore Integration Points — Analyze site code and data model to identify tables needing Web API integration
3. Review Integration Plan — Present findings to the user and confirm which tables to integrate
4. Implement Integrations — Use the webapi-integration agent for each table
5. Verify Integrations — Validate all expected files exist and the project builds successfully
6. Setup Permissions & Settings — Choose permissions source (upload diagram or let the architects analyze), then configure table permissions and Web API site settings with case-sensitive validated column names
7. Review & Deploy — Ask the user to deploy the site and invoke /deploy-site if confirmed

Phase 1: Verify Site Exists

Goal: Locate the Power Pages project root and confirm that prerequisites are met

Actions:

1.1 Locate Project

Look for powerpages.config.json in the current directory or immediate subdirectories to find the project root. Use your file-search tool (e.g., Glob with patterns powerpages.config.json and */powerpages.config.json) rather than a shell-specific command.

If not found: Tell the user to create a site first with /create-site.

1.2 Read Existing Config

Read powerpages.config.json to get the site name.

1.3 Detect Framework

Read package.json to determine the framework (React, Vue, Angular, or Astro). See ${PLUGIN_ROOT}/references/framework-conventions.md for the full framework detection mapping.

1.4 Check for Data Model

Look for .datamodel-manifest.json to discover available tables:

**/.datamodel-manifest.json

If found, read it — this is the primary source for table discovery.

1.5 Check Deployment Status

Look for the .powerpages-site folder:

**/.powerpages-site

If not found: Warn the user that the permissions phase (Phase 6) will require deployment first. The integration code (Phases 2–5) can still proceed.

1.6 Detect AI-only read mode (skill-to-skill invocation)

Inspect $ARGUMENTS. If the text begins with the sentinel [AI-READ-ONLY], the caller is another skill (typically /add-ai-webapi) that has already analysed the site and decided which tables need Layer 1/2 prerequisites for AI summarization reads. Parse the following structured tokens out of $ARGUMENTS:

When the sentinel is present:

• Set an internal flag AI-only read mode = true that every downstream phase consults.
• Skip the Phase 3 interactive table confirmation and use the provided tables list verbatim (user has already confirmed in the caller).
• The Phase 4.1 webapi-integration prompt restricts operations to read-only (list + get by id).
• The Phase 6 Path B agent prompts apply the hardened AI-only posture documented in each agent's "AI-only read mode" section.
• The Phase 6 Path A script invocations use --read only for table permissions and omit primary keys / relationship Navigation Properties from Webapi/<table>/fields.
• No AskUserQuestion prompts are issued for Phase 3 or Phase 6.2 — the caller owns those decisions.
• Defer all git commits to the caller. Skip Phase 4.4 (git add -A && git commit) and Phase 6.5 (permissions/settings commit) entirely. The caller is batching changes into one or two commits at orchestrator-defined milestones; an unprompted commit here turns one logical change into three. Print the file lists you would have committed so the caller can reproduce them.
• Suppress the end-of-skill deploy prompt. Skip Phase 6.1 (deploy-now ask when .powerpages-site is missing — the caller has already gated on this), Phase 7.3 (final deploy ask), and Phase 7.4 (post-deploy notes). The caller owns the single end-of-orchestration deploy decision; nesting deploy prompts inside the delegation gives the user 2–3 redundant asks per run. Return the integration summary (Phase 7.2) without trailing deploy/notes.

When the sentinel is absent: proceed exactly as today (full CRUD, full interactive flow, auto-commit, deploy prompt). This is the regression guard — no human invocation changes behavior.

Output: Confirmed project root, framework, data model availability, deployment status, and (if sentinel present) parsed AI-read-only contract.

Phase 2: Explore Integration Points

Goal: Analyze the site code and data model to identify all tables needing Web API integration

Actions:

Use the Explore agent (via Task tool with agent_type: "explore") to analyze the site code and data model. The Explore agent should answer these questions:

2.1 Discover Tables

Ask the Explore agent to identify all Dataverse tables that need Web API integration by examining:

• .datamodel-manifest.json — List of tables and their columns
• src/**/*.{ts,tsx,js,jsx,vue,astro} — Source code files that reference table data, mock data, or placeholder API calls
• Existing /_api/ fetch patterns in the code
• TypeScript interfaces or types that map to Dataverse table schemas
• Component files that display or manipulate data from Dataverse tables
• Mock data files or hardcoded arrays that should be replaced with API calls
• TODO or FIXME comments mentioning API integration

Prompt for the Explore agent:

"Analyze this Power Pages SPA site and identify all Dataverse tables that need Web API integration. Check .datamodel-manifest.json for the data model, then search the source code for: mock data arrays, hardcoded data, placeholder fetch calls to /_api/, TypeScript interfaces matching Dataverse column patterns (publisher prefix like cr*_), TODO/FIXME comments about API integration, and components that display table data. For each table found, report: the table logical name, the entity set name (plural), which source files reference it, what operations are needed (read/create/update/delete), and whether an existing API client or service already exists in src/shared/ or src/services/. Also check if src/shared/powerPagesApi.ts already exists."

When AI-only read mode is active (Phase 1.6 flag set): append the following to the prompt above:

"The caller has specified AI-READ-ONLY mode for tables [tables from sentinel] (primary=[primary], expand-targets=[expand-targets]). Operations needed for every table in that list = read only. Do NOT report create/update/delete call sites or mock-data replacement candidates. For each $expand target, also report the columns the primary's code $selects on that expansion (these become the minimal fields list)."

2.2 Identify Existing Integration Code

The Explore agent should also report:

• Whether src/shared/powerPagesApi.ts (or equivalent API client) already exists
• Which tables already have service files in src/shared/services/ or src/services/
• Which tables already have type definitions in src/types/
• Any framework-specific hooks/composables already created

This avoids duplicating work that was already done.

2.3 Compile Integration Manifest

From the Explore agent's findings, compile a list of tables needing integration:

Output: Complete integration manifest listing all tables, their operations, referencing files, and existing service status

Phase 3: Review Integration Plan

Goal: Present the integration manifest to the user and confirm which tables to integrate

Actions:

3.1 Present Findings

Show the user:

1. The tables that were identified for Web API integration
2. For each table: which files reference it, what operations are needed
3. Whether a shared API client already exists or needs to be created
4. Any tables that were skipped (already have services)

3.2 Confirm Tables

🚦 Gate (plan · integrate-webapi:3.2.confirm-tables): Final say on which tables get Web API integration code (client, types, services, hooks).

Trigger: Explore agent surfaced candidate tables in Phase 3.1. Why we ask: Auto-selecting all tables can generate orphaned TypeScript files for tables the user never intended to expose via Web API. Cancel leaves: Nothing — no service/type/hook files written yet.

When AI-only read mode is active (Phase 1.6 flag set): skip this step entirely. Use the tables list parsed from the sentinel verbatim — the caller has already confirmed the selection with the user. Do not issue an AskUserQuestion.

Otherwise, use AskUserQuestion to confirm:

If the user selects specific tables or adds more, update the integration manifest accordingly.

Output: User-confirmed list of tables to integrate (or sentinel-supplied list in AI-only read mode)

Phase 4: Implement Integrations

Goal: Create Web API integration code for each confirmed table using the webapi-integration agent

Actions:

4.1 Invoke Agent Per Table

For each table, use the Task tool to invoke the webapi-integration agent at ${PLUGIN_ROOT}/agents/webapi-integration.md:

Prompt template for the agent:

"Integrate Power Pages Web API for the [Table Display Name] table.

• Table logical name: [logical_name]
• Entity set name: [entity_set_name]
• Operations needed: [read/create/update/delete]
• Framework: [React/Vue/Angular/Astro]
• Project root: [path]
• Source files referencing this table: [list of files]
• Data model manifest path: [path to .datamodel-manifest.json if available]

Create the TypeScript types, CRUD service layer, and framework-specific hooks/composables. Replace any mock data or placeholder API calls in the referencing source files with the new service."

When AI-only read mode is active (Phase 1.6 flag set): replace "Operations needed: [read/create/update/delete]" with Operations needed: read-only and append to the prompt:

"AI-only read integration. Do NOT emit create, update, or delete functions. Scaffold only list<Table> (paginated) and get<Table>ById in the service layer. Create the framework-specific read hook only (e.g. use<Table>() with items, isLoading, error, refetch; no mutation hooks). Do not wire mock-data replacements beyond what is needed for the AI summarization caller — the upstream skill will add the summarization service on top. The shared src/shared/powerPagesApi.ts client is still created if it does not already exist; the caller relies on it for the AI integration's read path."

4.2 Process First Table, Then Parallelize Remaining

The first table must be processed alone — it creates the shared powerPagesApi.ts client that all other tables depend on. After the first table completes and the shared client exists:

• Verify the shared API client was created at src/shared/powerPagesApi.ts
• Then invoke all remaining tables in parallel using multiple Task calls — each table creates independent files (its own types in src/types/, service in src/shared/services/, and hook/composable), so there are no conflicts

If there is only one table, this step is simply sequential.

4.3 Verify Each Integration

After each agent completes (or after all parallel agents complete), verify the output:

• Check that the expected files were created (types, service, hook/composable)
• Confirm the shared API client exists after the first table is processed
• Note any issues reported by the agent

4.4 Git Commit

Skip when AI-only read mode is active (Phase 1.6 flag set) — the caller batches commits. Print the file list this phase would have staged so the caller can reproduce, then move on.

Otherwise, after all integrations are complete, stage and commit:

git add -A
git commit -m "Add Web API integration for [table names]"

Output: Integration code created for all confirmed tables, verified and (in normal mode) committed

Phase 5: Verify Integrations

Goal: Validate that all expected integration files exist, imports are correct, and the project builds successfully

Actions:

5.1 Verify File Inventory

For each integrated table, confirm the following files exist:

• Type definition in src/types/ (e.g., src/types/product.ts)
• Service file in src/shared/services/ or src/services/ (e.g., productService.ts)
• Framework-specific hook/composable (e.g., src/shared/hooks/useProducts.ts for React, src/composables/useProducts.ts for Vue)

Also verify:

• Shared API client at src/shared/powerPagesApi.ts exists
• Each service file references /_api/ endpoints
• Each service file imports from the shared API client

5.2 Verify Build

Run the project build to catch any import errors, type errors, or missing dependencies:

npm run build

If the build fails, fix the issues before proceeding. Common issues:

• Missing imports between generated files
• Type mismatches between service and type definitions
• Framework-specific compilation errors

5.3 Present Verification Results

Present a table summarizing the verification:

Build status: Pass / Fail (with details)

Output: All integration files verified, project builds successfully

Phase 6: Setup Permissions & Settings

Goal: Configure table permissions and Web API site settings for all integrated tables using the table-permissions-architect and webapi-settings-architect agents

Actions:

6.1 Check Deployment Prerequisite

Both agents require the .powerpages-site folder.

When AI-only read mode is active (Phase 1.6 flag set): the caller has already gated on .powerpages-site existing — re-check it as a guard, and if it is genuinely absent, stop and report the contract violation back to the caller (the caller will surface this to the user). Do NOT issue the deploy prompt below — the caller owns the single deploy decision.

Otherwise, if .powerpages-site doesn't exist:

🚦 Gate (plan · integrate-webapi:6.1.deploy-first): .powerpages-site missing — needed by both architect agents. Deploy first, or skip permissions setup and finish without them.

Trigger: Phase 6.1 found no .powerpages-site folder. Why we ask: Auto-skipping leaves the integration broken (no permissions to back the Web API calls); auto-deploying picks the wrong env. Cancel leaves: Nothing — services/types/hooks from Phase 4 stay on disk regardless.

Use AskUserQuestion:

If "Yes, deploy now": Invoke /deploy-site first, then resume this phase.

If "Skip": Skip to Phase 7 with a note that permissions and site settings still need to be configured.

6.2 Choose Permissions Source

🚦 Gate (plan · integrate-webapi:6.2.permissions-source): Decide between uploading an existing permissions diagram (Path A) and letting the architect agents derive it (Path B). Choice routes the rest of Phase 6.

Trigger: Entering Phase 6.2 after deployment prerequisite is satisfied. Why we ask: Path A produces table permissions matching a stale or wrong diagram; Path B can take minutes to query Dataverse. Cancel leaves: Nothing — no permission YAML written yet.

When AI-only read mode is active (Phase 1.6 flag set): skip the permissions-source question entirely and default to Path B (let the architects figure it out) — proceed directly to section 6.3. Per the Phase 1.6 contract, no AskUserQuestion is issued here; the caller (/add-ai-webapi) owns this decision, and the AI-only architect prompts in 6.3 already encode the read-only posture. Path A (upload an existing permissions diagram) is an interactive human path and does not apply to delegated runs.

Otherwise, ask the user how they want to define the permissions using the AskUserQuestion tool:

Question: "How would you like to define the Web API permissions and settings for your site?"

Route to the appropriate path:

Path A: Upload Existing Permissions Diagram

🚦 Gate (plan · integrate-webapi:6.2.permissions-approval): Final sign-off on the parsed permissions plan (from the uploaded diagram) before any web-role / table-permission / site-setting YAML write. Fires at step 6 of the Path A sequence below.

Trigger: Path A — diagram parsed and Mermaid flowchart rendered. Why we ask: Wrong scope / wrong CRUD flags get committed to .powerpages-site/table-permissions/ — fixable but noisy in git history. Cancel leaves: Nothing — no YAML files written yet.

If the user chooses to upload an existing diagram:

1. Ask the user to provide their permissions diagram. Supported formats:

   • Image file (PNG, JPG) — Use the Read tool to view the image and extract web roles, table permissions, CRUD flags, scopes, and site settings from it
   • Mermaid syntax — The user can paste a Mermaid flowchart diagram text directly in chat
   • Text description — A structured list of web roles, table permissions, scopes, and site settings

2. Parse the diagram into structured format:

   • Web roles: Match with existing roles from .powerpages-site/web-roles/ by name to get their UUIDs
   • Table permissions: Permission name, table logical name, web role UUID(s), scope, CRUD flags (read/create/write/delete/append/appendto), parent permission and relationship name (if Parent scope)
   • Site settings: Webapi/<table>/enabled and Webapi/<table>/fields — CRITICAL: fields must list every required column LogicalName explicitly, including columns used by aggregate OData queries

3. Validate column names against Dataverse — Even when using a user-provided diagram, query Dataverse for each table's column LogicalNames and verify that every column in the Webapi/<table>/fields values uses the exact Dataverse LogicalName (case-sensitive). Correct any mismatches before creating files.

4. Cross-check with existing configuration in .powerpages-site/ to identify which permissions and site settings are new vs. already exist.

5. Generate a Mermaid flowchart from the parsed data (if the user provided an image or text) for visual confirmation.

6. Present the parsed permissions plan to the user for approval using AskUserQuestion:

   Question	Options
   Does this permissions plan look correct?	Approve and create files (Recommended), Request changes, Cancel

7. Proceed directly to section 6.4: Create Permission & Settings Files with the parsed data.

Path B: Let the Architects Figure It Out

If the user chooses to let the architects figure it out, proceed to section 6.3: Invoke Table Permissions Agent.

6.3 Invoke Table Permissions and Web API Settings Agents (in Parallel)

These two agents are independent — invoke them in parallel using two Task calls simultaneously:

Table Permissions Agent

Use the Task tool to invoke the table-permissions-architect agent at ${PLUGIN_ROOT}/agents/table-permissions-architect.md:

Prompt (default — full CRUD):

"Analyze this Power Pages SPA site and propose table permissions. The following tables have been integrated with Web API: [list of tables integrated in Phase 4]. Check for existing web roles and table permissions. Propose a complete table permissions plan covering all integrated tables. After I approve the plan, create the web role and table permission YAML files using the deterministic scripts."

Prompt (AI-only read mode active): append to the default prompt:

"AI-only read integration. The caller is /add-ai-webapi and these tables will be summarised by /_api/summarization/data/v1.0/ — the endpoint is semantically a read and never mutates Dataverse. Therefore:

• Set read: true only. Do NOT propose create, write, or delete even if the default convention would include them.
• For collection-valued $expand targets (primary=[primary], expand-targets=[expand-targets]), use Parent scope with read: true, and add appendTo: true on the parent table permission so the relationship traversal is allowed.
• If no suitable web role exists, surface this back to the caller — /add-ai-webapi will invoke /create-webroles up front; do not block on role creation here."

The agent will:

1. Analyze the site and propose a plan (with Mermaid diagram)
2. Present the plan via plan mode for user approval
3. After approval, create any needed web roles using create-web-role.js
4. Create all table permission files using create-table-permission.js
5. Return a summary of created files

Web API Settings Agent

Use the Task tool to invoke the webapi-settings-architect agent at ${PLUGIN_ROOT}/agents/webapi-settings-architect.md:

Prompt (default — full CRUD):

"Analyze this Power Pages SPA site and propose Web API site settings. The following tables have been integrated with Web API: [list of tables integrated in Phase 4]. Check for existing site settings and query Dataverse for exact column LogicalNames. Propose site settings with case-sensitive validated column names. After I approve the plan, create the site setting YAML files using the deterministic scripts."

Prompt (AI-only read mode active): append to the default prompt:

"AI-only read integration. These tables will be summarised by /_api/summarization/data/v1.0/. The fields list rules are stricter than the default CRUD posture:

• Webapi/<table>/fields must contain exactly the columns required by every site call against the table. Start with the primary's $select / $expand; if the table is also used by an aggregate query, add its grouping keys, aggregate inputs, filter columns, and ordered columns.
• Do not include the primary key column. The summarization endpoint carries the record id in the URL path; Microsoft's shipped case preset ships Webapi/incident/fields = description,title with no incidentid.
• For lookup columns, include only the _<col>_value OData read form. Add a relationship Navigation Property only when non-AI code uses it before @odata.bind.
• Still query Dataverse for exact LogicalNames (case-sensitive) — case mismatches produce 403."

The agent will:

1. Analyze the site, query Dataverse for exact column LogicalNames
2. Cross-validate column names (case-sensitive)
3. Present the plan via plan mode for user approval
4. After approval, create all site setting files using create-site-setting.js
5. Return a summary of created files

Wait for both agents to complete before proceeding to 6.4.

6.4 Create Permission & Settings Files (Path A Only)

This section applies only to Path A (user-provided permissions diagram). For Path B, the architect agents create the files directly in section 6.3.

After parsing the user's diagram, create the YAML files using the deterministic scripts below. Do NOT write YAML files manually — always use these scripts which handle UUID generation, field ordering, formatting, and file naming automatically.

6.4.1 Create Web Roles (if needed)

If the plan requires new web roles that don't already exist, create them first (their UUIDs are needed for table permissions):

node "${PLUGIN_ROOT}/skills/create-webroles/scripts/create-web-role.js" --projectRoot "<PROJECT_ROOT>" --name "<Role Name>" [--anonymous] [--authenticated]

Capture the JSON output ({ "id": "<uuid>", "filePath": "<path>" }) — use the id as the --webRoleIds value when creating table permissions.

6.4.2 Create Table Permissions

For each table permission in the plan. Process parent permissions before child permissions — children need the parent's UUID from the JSON output.

When AI-only read mode is active (Phase 1.6 flag set): the default flag set is --read only. Do NOT pass --create, --write, or --delete for any permission in scope — AI summarization reads never mutate. For Parent-scope child permissions, still pass --parentPermissionId and --parentRelationshipName; on the parent permission add --appendto so the AI endpoint can traverse the relationship.

For Global/Contact/Account/Self scope:

node "${PLUGIN_ROOT}/scripts/create-table-permission.js" --projectRoot "<PROJECT_ROOT>" --permissionName "<Permission Name>" --tableName "<table_logical_name>" --webRoleIds "<uuid1,uuid2>" --scope "Global" [--read] [--create] [--write] [--delete] [--append] [--appendto]
node "${PLUGIN_ROOT}/scripts/create-table-permission.js" --projectRoot "<PROJECT_ROOT>" --permissionName "<Permission Name>" --tableName "<table_logical_name>" --webRoleIds "<uuid1,uuid2>" --scope "Contact" --contactRelationshipName "<lookup_to_contact>" [--read] [--create] [--write] [--delete] [--append] [--appendto]
node "${PLUGIN_ROOT}/scripts/create-table-permission.js" --projectRoot "<PROJECT_ROOT>" --permissionName "<Permission Name>" --tableName "<table_logical_name>" --webRoleIds "<uuid1,uuid2>" --scope "Account" --accountRelationshipName "<lookup_to_account>" [--read] [--create] [--write] [--delete] [--append] [--appendto]
node "${PLUGIN_ROOT}/scripts/create-table-permission.js" --projectRoot "<PROJECT_ROOT>" --permissionName "<Permission Name>" --tableName "<table_logical_name>" --webRoleIds "<uuid1,uuid2>" --scope "Self" [--read] [--create] [--write] [--delete] [--append] [--appendto]

For Parent scope (requires parent permission UUID and relationship name):

node "${PLUGIN_ROOT}/scripts/create-table-permission.js" --projectRoot "<PROJECT_ROOT>" --permissionName "<Permission Name>" --tableName "<table_logical_name>" --webRoleIds "<uuid1>" --scope "Parent" --parentPermissionId "<parent-uuid>" --parentRelationshipName "<relationship_name>" [--read] [--create] [--write] [--delete] [--append] [--appendto]

Each invocation outputs { "id": "<uuid>", "filePath": "<path>" }. Use the id as --parentPermissionId for child permissions.

6.4.3 Create Site Settings

For each site setting in the plan:

Enabled setting (boolean):

node "${PLUGIN_ROOT}/scripts/create-site-setting.js" --projectRoot "<PROJECT_ROOT>" --name "Webapi/<table>/enabled" --value "true" --description "Enable Web API access for <table> table" --type "boolean"

Fields setting (string — use the validated column names from the diagram):

node "${PLUGIN_ROOT}/scripts/create-site-setting.js" --projectRoot "<PROJECT_ROOT>" --name "Webapi/<table>/fields" --value "<comma-separated-validated-columns>" --description "Allowed fields for <table> Web API access"

Inner error setting (boolean, optional for debugging):

node "${PLUGIN_ROOT}/scripts/create-site-setting.js" --projectRoot "<PROJECT_ROOT>" --name "Webapi/error/innererror" --value "true" --description "Enable detailed error messages for debugging" --type "boolean"

Important: The --value for every fields setting MUST follow ${PLUGIN_ROOT}/references/webapi-field-allowlist.md. Use exact LogicalNames for ordinary columns, _<LogicalName>_value for lookup reads, and the exact Navigation Property used before @odata.bind for lookup writes. Wildcard field access is unsupported beginning September 14, 2026.

Aggregate queries: Include every grouping key, aggregate input, filter column, and ordered column referenced by $apply in the explicit fields list.

Lookup columns: Include _<LogicalName>_value for reads. If POST/PATCH sets the relationship with NavigationProperty@odata.bind, also include that exact case-sensitive Navigation Property in the fields value.

AI-only read mode (Phase 1.6 flag set): the fields-value rules tighten:

• Include the exact columns required by every site call against the table. Start with the primary's $select / $expand; if the table is also used by an aggregate query, add its grouping keys, aggregate inputs, filter columns, and ordered columns.
• Do NOT include the primary key. The summarization endpoint carries the record id in the URL path; Microsoft's shipped case preset ships Webapi/incident/fields = description,title with no incidentid.
• For lookup columns, include only the _<col>_value read form. Omit relationship Navigation Properties because AI mode has no @odata.bind writes.
• Keep File/Image and aggregate-query columns explicit, adding only the columns the site actually uses.

6.5 Git Commit

Skip when AI-only read mode is active (Phase 1.6 flag set) — the caller batches commits. Print the file list this phase would have staged so the caller can reproduce, then move on.

Otherwise, stage and commit the permission and settings files:

git add -A
git commit -m "Add table permissions and Web API site settings for [table names]"

Output: Table permissions and site settings created, verified, and (in normal mode) committed

6.6 Final site-integrity gate

After all integration source changes, permission/settings work, and build verification are complete, run:

node "${PLUGIN_ROOT}/scripts/validate-site-integrity.js" --projectRoot "<PROJECT_ROOT>"

Fix every blocking error before presenting completion or asking to deploy. Inspect and report review findings in both directions and with expanded content. Rerun this gate after any later source change.

Phase 7: Review & Deploy

Goal: Present a summary of all work performed and offer deployment

Actions:

7.1 Record Skill Usage

Reference: ${PLUGIN_ROOT}/references/skill-tracking-reference.md

Follow the skill tracking instructions in the reference to record this skill's usage. Use --skillName "IntegrateWebApi".

7.2 Present Summary

Present a summary of everything that was done:

7.3 Ask to Deploy

🚦 Gate (plan · integrate-webapi:7.3.deploy): Post-integration deploy prompt — Web API calls won't work until permissions and site settings are deployed.

Trigger: All integration code + permissions YAML committed. Why we ask: Auto-deploy picks whatever env PAC CLI happens to point at. Cancel leaves: Nothing — integration artifacts stay on disk; no deploy fired.

Skip when AI-only read mode is active (Phase 1.6 flag set) — the caller owns the single end-of-orchestration deploy decision. Return the Phase 7.2 summary and stop.

Otherwise, use AskUserQuestion:

If "Yes, deploy now": Invoke the /deploy-site skill to deploy the site.

If "No, I'll deploy later": Acknowledge and remind:

"No problem! Remember to deploy your site using /deploy-site when you're ready. The Web API calls will not work until the site is deployed with the new permissions."

7.4 Post-Deploy Notes

Skip when AI-only read mode is active — the caller emits its own post-deploy guidance that covers AI-specific concerns (governance hierarchy, runtime version, Bing dependency).

Otherwise, after deployment (or if skipped), remind the user:

• Test the API: Open the deployed site and verify Web API calls work in the browser's Network tab
• Check permissions: If any API call returns 403, verify table permissions and site settings are correct. The most common cause is a column in the request missing from Webapi/<table>/fields or not matching the exact case-sensitive Dataverse LogicalName returned by metadata. For aggregate OData ($apply, aggregate, grouped totals), verify that every grouping key, aggregate input, filter column, and ordered column is listed explicitly.
• Disable innererror in production: If Webapi/error/innererror was enabled for debugging, disable it before going live
• Web roles: Users must be assigned the appropriate web roles to access protected APIs

Output: Summary presented; deployment completed or deferred (normal mode), or returned to caller (AI-only read mode).

Important Notes

Throughout All Phases

• Use TaskCreate/TaskUpdate to track progress at every phase
• Ask for user confirmation at key decision points (see list below)
• First table sequential, then parallel — the first table creates the shared API client; after that, remaining tables can be processed in parallel since each creates independent files
• Commit at milestones — after integration code and after permission files
• Verify each integration — confirm expected files exist after each agent invocation

Key Decision Points (Wait for User)

1. After Phase 2: Confirm which tables to integrate
2. After Phase 3: Approve integration plan
3. At Phase 6.1: Deploy now or skip permissions (if .powerpages-site missing)
4. At Phase 6.2: Choose permissions source (upload diagram or let the architects analyze)
5. At Phase 6.3: Approve table permissions plan and Web API site settings plan (both agents run in parallel for Path B, each presents its own plan for approval)
6. At Phase 7.2: Deploy now or deploy later

Progress Tracking

Before starting Phase 1, create a task list with all phases using TaskCreate:

Mark each task in_progress when starting it and completed when done via TaskUpdate. This gives the user visibility into progress and keeps the workflow deterministic.

Begin with Phase 1: Verify Site Exists