Migrate Webapi Selectall

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

Migrate Power Pages Web API Wildcards

Replace every deprecated Webapi/<table>/fields = * value with the smallest explicit column set proven by the site's actual Web API behavior.

The LLM owns source discovery, call-chain reasoning, field decisions, report writing, and edits. Use the bundled script only to retrieve authoritative Dataverse table schema; it must not decide which columns the code needs.

Support both:

• traditional sites with HTML, JavaScript, Liquid, web templates, and aggregate YAML;
• SPA sites with React, Vue, Angular, Astro, TypeScript, downloaded deployment YAML, or mixed custom JavaScript.

Initial request: $ARGUMENTS

Non-negotiable rules

1. Review every discovered source table Web API call, including shared wrappers, dynamic builders, and response consumers.
2. Review every configuration scope and deployment-profile copy.
3. Map request EntitySetName values to setting logical names using table schema. Never singularize, pluralize, or guess.
4. Treat * as unsupported for reads, writes, aggregates, FetchXML, files, and images.
5. Give every wildcard an exact proposed replacement before editing anything.
6. Report every already-explicit fields setting, including missing and potentially unnecessary columns.
7. Never apply a partial wildcard plan. Resolve all wildcards and call-site rows first.
8. Keep reports free of absolute local paths, tokens, URLs, data values, filter literals, request bodies, response bodies, and source snippets, and build them only by rendering the bundled template with the bundled script.
9. Preserve unrelated YAML structure and values.
10. Verify with a fresh discovery pass, not remembered inventory.
11. Leave only the rendered report and its icon in the migration output directory.
12. Never download or upload site content until the user has explicitly confirmed the environment, website, site type, data model, and deployment profile. Neither transfer can be reverted.
13. Run smoke tests only after explicit approval, and never issue POST, PATCH, PUT, or DELETE Web API calls against a deployed site. Testing a write destroys real record data.

Read references/column-analysis.md before analyzing calls. Read references/configuration-and-reporting.md before inventorying settings or writing the report. Read references/site-transfer.md before any download or upload.

Phase 1: Prepare

Goal: Resolve the project, confirm the site, and protect existing work.

1. Create all seven tasks from Progress tracking.
2. Resolve PROJECT_ROOT from $ARGUMENTS or the current directory.
3. Detect site markers independently:
   • powerpages.config.json indicates an SPA site;
   • root website.yml, root sitesetting.yml, or .powerpages-site/ indicates downloaded declarative artifacts;
   • when both appear, scan both layouts.
4. Read .solution-manifest.json when present. This migration changes existing settings; do not create or select another solution.
5. Inspect git status. Never discard, hide, or include unrelated user changes.
6. Run node --version.
7. Confirm assets/migration-report-template.html and scripts/render-migration-report.js are readable, and stop if either is missing.
8. Read references/site-transfer.md, then confirm the environment, website name and WebSiteId, site type, data model, deployment profile, and target path with the user. Check each against pac auth who, pac env who, and pac pages list, and stop on any mismatch. Never infer one from a folder name or an active default.

Analyzing the wrong site produces confident, wrong fixes, so settle identity before reading any setting. Download only when the user wants a fresh copy or PROJECT_ROOT holds no site content; downloading replaces local files and cannot be reverted.

<!-- gate: migrate-webapi-selectall:1.download-site | category=consent | cancel-leaves=nothing -->

🚦 Gate (consent · migrate-webapi-selectall:1.download-site): Approve the download only after displaying the confirmed environment, website name and ID, site type, data model, target path, and the exact command. Canceling leaves local content untouched and continues against the existing copy.

Use AskUserQuestion: Download the confirmed site or Use the local copy. Repeat step 3 after any download.

Output: Project root, site layouts, solution context, git state, confirmed site identity, and a downloaded copy when approved.

Phase 2: Build the complete inventory

Goal: Find every configuration and candidate source Web API call before reasoning about columns.

2.1 Inventory configuration scopes

Use Glob, Grep, and Read to inspect:

• every sitesetting.yml;
• every *.sitesetting.yml;
• .powerpages-site/site-settings/;
• deployment-profile and environment-specific copies.

Record every Webapi/<table>/fields and Webapi/<table>/enabled entry with its relative file, line, scope, key style, and current value. Classify fields settings as:

• wildcard;
• explicit;
• missing for an enabled table;
• duplicate only within the same configuration scope.

Do not treat identical settings in different deployment profiles as duplicates. Record every profile name and which settings it overrides. Never assume the default profile is the intended deployment profile.

Query Dataverse once per unique table, never once per configuration.

2.2 Inventory source calls

Analyze only authoritative, editable source files. Never inspect compiled or generated code.

For SPA sites:

1. Read powerpages.config.json, package.json, and present framework or bundler configuration before searching calls.
2. Treat compiledPath and every configured build-output directory as a hard exclusion.
3. Exclude .powerpages-site/web-files/, node_modules/, coverage and cache directories, source maps, minified bundles, framework output directories, and content-hashed assets matching <entry-name>-<content-hash>.<extension>.
4. Use .powerpages-site/site-settings/ only for configuration inventory, never for source analysis.
5. Search editable roots such as configured source directories and framework application directories.

For traditional sites, search editable JavaScript, Liquid, web templates, web files, and other authored source. Do not exclude an authored traditional web file merely because it is deployed as a web file.

Search source extensions including .js, .jsx, .ts, .tsx, .vue, .html, .htm, .liquid, .aspx, .ascx, .cshtml, and XML web templates. If a call exists only in compiled, minified, generated, or content-hashed output, record a missing-source blocker and stop the migration. Do not infer columns from that output.

Search for:

• /_api/, encoded variants, split URL fragments, and API base constants;
• fetch, Axios, XMLHttpRequest, jQuery AJAX, webapi.safeAjax, shell.ajaxSafePost, and custom request wrappers;
• entity-set constants, query builders, FetchXML builders, and body builders;
• callers and consumers imported from other files.

For each source candidate, record relative path, line, method, endpoint expression, and wrapper chain. A comment, example, or non-table endpoint still needs an explicit disposition. Excluded build outputs are never recorded.

Write the settings inventory to docs/webapi-selectall-migration/migration-report.json and render the draft report as described in references/configuration-and-reporting.md. Do not propose fields yet.

Stop if any in-scope source or configuration file cannot be read.

<!-- gate: migrate-webapi-selectall:2.confirm-scope | category=plan | cancel-leaves=draft-migration-report -->

🚦 Gate (plan · migrate-webapi-selectall:2.confirm-scope): Confirm the project, configuration scopes and profiles, wildcard count, explicit-setting count, and source inventory before schema retrieval. Canceling leaves only the read-only draft report.

Use AskUserQuestion to confirm or cancel. Expand the inventory and repeat this phase if the user identifies another source or deployment scope.

Phase 3: Retrieve schema and resolve columns

Goal: Use authoritative names while letting the LLM determine actual usage.

3.1 Retrieve only relevant table schema

Build the initial unique list containing:

• logical table names from all Web API settings;
• entity-set names from all candidate table calls;
• entity-set names directly present in bind targets or related-table calls.

Do not treat a navigation-property name as a table identifier. Its target logical name is authoritative only after relationship metadata resolves it.

Resolve the environment URL from confirmed project context or pac env who. If unavailable, ask for the URL as data gathering; never ask for or accept an access token.

<!-- not-a-gate: environment URL supplies read-only metadata query input only -->

Write every deduplicated identifier, one per line, to docs/webapi-selectall-migration/table-identifiers.txt. Run:

node "${PLUGIN_ROOT}/skills/migrate-webapi-selectall/scripts/query-table-schema.js" --project-root "<PROJECT_ROOT>" --environment-url "<ENVIRONMENT_URL>" --tables-file "<PROJECT_ROOT>/docs/webapi-selectall-migration/table-identifiers.txt" --output "<PROJECT_ROOT>/docs/webapi-selectall-migration/table-schema.json"

If an identifier does not resolve, trace the code or obtain the correct contract; do not guess.

After the initial snapshot:

1. Match every used $expand navigation property against its source table's returned relationship metadata.
2. Collect only target logical names absent from all existing snapshots.
3. Write those names to table-identifiers-pass-<N>.txt and query them to table-schema.pass-<N>.json.
4. Repeat for nested expansion paths until every used navigation segment is resolved.

Treat table-schema.json and all numbered snapshots as one schema package. Never requery a logical table already present in that package, and never launch concurrent schema queries.

Identifier lists and schema snapshots are working files. Keep them until verification finishes, then delete them in Phase 7.

3.2 Analyze every call and consumer

For every source inventory row:

1. Read the complete enclosing function or template block.
2. Trace imported wrappers, URL variables, query builders, body builders, response mappers, types, components, templates, and every caller.
3. Follow conditional branches, spreads, dynamic arrays, and runtime configuration.
4. Map the entity set to its logical table using the schema package.
5. Apply every rule in references/column-analysis.md.
6. Record requirements per owning logical table, not merely per request.
7. Classify the row as mapped, non-table, or not-a-call.
8. Leave the row unresolved if any runtime branch or consumer remains unknown.

One logical table is normally reached from several places. Collect every call site for a table before proposing its fields: duplicated or competing wrappers, per-page scripts, different query shapes, and repeated calls in the same file. A later call site for an already-analyzed table can still add columns, so never stop at the first one.

For normal record GETs without $select, derive output fields from every consumer and propose a source edit adding the smallest explicit projection. Filters, ordering, and other query fields still belong in the fields setting, even when they should not be added to the output projection.

Keep the response projection and fields-setting allowlist as separate sets. Never add a filter-only, order-only, grouping-only, or write-only column to an existing $select unless a response consumer also reads it.