Add agent dev skill for search

.github/skills/find-package-skill/SKILL.md

@@ -0,0 +1,25 @@
+---
+name: find-package-skill
+description: '**UTILITY SKILL** — Must be consulted before making any changes to packages under sdk/. Provides package-specific domain knowledge that prevents common mistakes. WHEN: "add feature to package", "fix bug in package", "modify package code", "regenerate from typespec", "change client library".'
+---
+
+# Find Package Skill
+
+Before modifying any SDK package in this monorepo, check this registry to see
+if a package-specific skill exists. Package skills contain tribal knowledge
+(architecture, data flows, type mappings, pitfalls) that prevents common mistakes.
+
+Always check this registry before modifying any SDK package — even if you think
+you already know the package well.
+
+## How to Use
+
+1. Find the package you're modifying in the table below.
+2. Read the SKILL.md at the listed path using the Read tool. Then read all files under the `references/` directory next to it for additional context.
+3. If the package isn't listed, no package-specific skill exists yet — proceed normally.
+
+## Package Skills
+
+| Package                   | Path                                                                        |
+| ------------------------- | --------------------------------------------------------------------------- |
+| `azure-search-documents` | `sdk/search/azure-search-documents/.github/skills/search-documents/SKILL.md` |

sdk/search/azure-search-documents/.github/skills/search-documents/SKILL.md

@@ -0,0 +1,224 @@
+---
+name: search-documents-python
+description: 'Post-regeneration customization guide for the azure-search-documents Python SDK. After running tsp-client update, consult this skill to re-apply all search-specific customizations and produce a production-ready SDK. USE WHEN: regenerating the search SDK from TypeSpec, running tsp-client update, fixing broken _patch.py imports after regeneration, adding a new operation or model to the search SDK, verifying the SDK still works after spec changes, or any task that mentions "regenerate", "tsp-client", "typespec", or "codegen" in the context of this package.'
+---
+
+# azure-search-documents — Post-Regeneration Customization Guide
+
+After running `tsp-client update`, the generated code is a raw skeleton — not a shippable SDK. This skill tells you exactly what customizations exist in `_patch.py` files and how to verify they still work after regeneration.
+
+The generator never touches `_patch.py` files, so your customizations survive regeneration. But they import from generated modules that **do** change. A renamed model, a new parameter, or a removed enum value will silently break a `_patch.py` file. The verification steps below catch these breaks.
+
+## Step 1: Run Regeneration
+
+```bash
+cd sdk/search/azure-search-documents
+
+# Update tsp-location.yaml with the new spec commit SHA, then:
+tsp-client update
+
+# If API version changed, update _metadata.json too
+```
+
+## Step 2: Verify Imports in All `_patch.py` Files
+
+Every `_patch.py` with customizations imports from generated modules. After regeneration, check that these imports still resolve. The fastest way:
+
+```bash
+python -c "from azure.search.documents import SearchClient"
+python -c "from azure.search.documents.aio import SearchClient"
+python -c "from azure.search.documents.indexes.models import SearchField, SearchFieldDataType"
+python -c "from azure.search.documents.models import IndexDocumentsBatch"
+```
+
+If any fail, a generated class/enum was renamed or removed. Read `references/customizations.md` for the exact import each `_patch.py` depends on.
+
+The `_patch.py` files that have customizations (all others are empty boilerplate):
+
+```
+azure/search/documents/
+├── _patch.py                    # SearchClient, SearchIndexingBufferedSender, ApiVersion
+├── _operations/_patch.py        # SearchItemPaged, search(), document CRUD
+├── models/_patch.py             # IndexDocumentsBatch, RequestEntityTooLargeError
+├── aio/_patch.py                # Async SearchClient, async SearchIndexingBufferedSender
+├── aio/_operations/_patch.py    # AsyncSearchItemPaged, async search()/CRUD
+├── indexes/_operations/_patch.py        # Polymorphic delete/update, list helpers
+├── indexes/models/_patch.py             # SearchField, field builders, enum aliases
+└── indexes/aio/_operations/_patch.py    # Async index/indexer operations
+```
+
+## Step 3: Check ApiVersion
+
+After regeneration, the generated code targets the API version in `_metadata.json`. The hand-maintained `ApiVersion` enum in `_patch.py` must include this version.
+
+```bash
+# 1. See what API version the generator used
+python -c "import json; print(json.load(open('_metadata.json'))['apiVersion'])"
+
+# 2. See what versions the SDK currently advertises
+grep -A 20 'class ApiVersion' azure/search/documents/_patch.py
+
+# 3. Check the current default
+grep 'DEFAULT_VERSION' azure/search/documents/_patch.py
+```
+
+If the generated API version is **not** in the `ApiVersion` enum:
+1. Add the new member to `ApiVersion` in `azure/search/documents/_patch.py` (e.g., `V2026_05_01_PREVIEW = "2026-05-01-preview"`)
+2. Update `DEFAULT_VERSION` to point to the new member
+3. Update the `ApiVersion` docstring in the class if one exists
+
+Verify the default round-trips correctly:
+
+```bash
+python -c "from azure.search.documents._patch import ApiVersion, DEFAULT_VERSION; print(DEFAULT_VERSION.value)"
+```
+
+## Step 4: Check for New Operations That Need Wrappers
+
+Look at what changed in the generated operations files:
+
+```bash
+git diff --name-only | grep "_operations\.py" | grep -v _patch
+```
+
+If a new operation was added to the generated `_operations.py`, decide whether it needs a convenience wrapper in `_patch.py`. Operations that need wrappers:
+- **Delete operations** — need the polymorphic str-or-model pattern (see "Polymorphic Delete Pattern" below)
+- **Create-or-update operations** — need `prefer="return=representation"`, `match_condition`, and `etag` forwarding
+- **List operations** — may need a `select` parameter or name-only projection
+
+Operations that pass through without a wrapper (if the generated signature is already user-friendly) can be left alone — they're inherited from the generated mixin.
+
+Remember: every sync wrapper in `_operations/_patch.py` needs a matching async wrapper in `aio/_operations/_patch.py`.
+
+## Step 5: Check for Model/Enum Changes That Affect Customizations
+
+```bash
+git diff models/_enums.py models/_models.py indexes/models/_enums.py indexes/models/_models.py
+```
+
+Watch for:
+- **Renamed enum values** — backward-compat aliases in `_patch.py` may reference old names
+- **New enum values that collide with Python keywords** — the generator renames them with `_ENUM` suffix (e.g., `IS` → `IS_ENUM`), and you must add a backward-compat alias
+- **Changed model constructors** — `IndexDocumentsBatch`, `SearchField`, `SearchIndexerDataSourceConnection`, `KnowledgeBase` are subclassed in `_patch.py` and may break if the base constructor changes
+- **New fields on `SearchResult`** — `_convert_search_result()` in `_operations/_patch.py` extracts `@search.*` metadata fields; new ones need to be added
+- **Changed `SearchRequest` model** — `_build_search_request()` constructs this model directly; new parameters need to be wired through
+- **Changed `SearchFieldDataType` model** — `indexes/models/_patch.py` monkey-patches `SearchFieldDataType.Collection` as a `staticmethod` and adds camelCase backward-compat aliases (e.g., `Int32` → `INT32`). If values are added, removed, or renamed in the generated enum, the aliases and `Collection` helper must be updated to match
+
+## Step 6: Ensure mypy pass
+
+```bash
+cd sdk/search/azure-search-documents
+
+# Preferred — uses azpysdk CLI (install via: pip install -e eng/tools/azure-sdk-tools)
+azpysdk mypy
+```
+
+The package-level `mypy.ini` already ignores `_generated` internals — you only need to fix errors in `_patch.py` files and `samples/`.
+
+## Step 7: Ensure pylint pass
+
+```bash
+cd sdk/search/azure-search-documents
+
+# Preferred
+azpysdk pylint
+```
+
+The repo-level `pylintrc` already excludes `_generated/`, `_vendor/`, `tests/`, and `samples/` — only `_patch.py` customizations are linted.
+
+## Step 8: Update Documentation and Samples
+
+If the regeneration added new operations, models, or changed the API version, update the docs and samples:
+
+### Changelog
+
+```bash
+# Edit CHANGELOG.md — add an entry under the next unreleased version
+# Format:
+#   ## 11.x.0bN (Unreleased)
+#   ### Features Added
+#   - Added `new_operation` to `SearchIndexClient`
+#   ### Breaking Changes
+#   - Renamed `OldModel` to `NewModel`
+code CHANGELOG.md
+```
+
+### README
+
+If new client classes or major features were added, update `README.md` with usage examples:
+
+```bash
+code README.md
+```
+
+# Customization Patterns Reference
+
+Each pattern below describes a customization that exists in the codebase today. After regeneration, verify each one still works. Read `references/customizations.md` for the exhaustive file-by-file inventory.
+
+## SearchClient Constructor Reordering
+
+`SearchClient` in `_patch.py` swaps parameter order to `(endpoint, index_name, credential)`. The generated base uses `(endpoint, credential, index_name)`. If regeneration changes the base constructor signature, update the subclass.
+
+## Custom Search Pagination
+
+`search()` does **not** use standard Azure SDK paging. It uses `SearchItemPaged` / `AsyncSearchItemPaged` with a custom `SearchPageIterator` because:
+- Search paginates via POST with `nextPageParameters` body, not GET with `nextLink`
+- First-page metadata (facets, count, answers) must be available before iteration
+- Results are converted to dicts with `@search.*` keys
+
+The continuation token is Base64-encoded JSON: `{"apiVersion": "...", "nextLink": "...", "nextPageParameters": {...}}`
+
+Shared helpers (`_build_search_request`, `_convert_search_result`, `_pack_continuation_token`, `_unpack_continuation_token`) live in sync `_operations/_patch.py` and are imported by async. Don't duplicate them.
+
+## Pipe-Delimited Semantic Encoding
+
+`_build_search_request()` encodes semantic parameters into pipe-delimited wire format:
+```
+query_answer="extractive", count=3  →  "extractive|count-3"
+query_caption="extractive", highlight=True  →  "extractive|highlight-true"
+```
+New semantic parameters should follow this pattern.
+
+## SearchIndexingBufferedSender
+
+Entirely hand-authored in `_patch.py` (sync) and `aio/_patch.py` (async). Not generated. Wraps `SearchClient` with auto-flush timer, 413 recursive batch splitting, retry per document key (409/422/503), and key field auto-detection. The async version supports both sync and async callbacks via `asyncio.iscoroutinefunction()`.
+
+## Field Builders
+
+`SimpleField()`, `SearchableField()`, `ComplexField()` in `indexes/models/_patch.py`. `SimpleField` explicitly sets `searchable=False`. `SearchableField` auto-sets type to `String`/`Collection(String)`.
+
+`SearchField` subclass adds `hidden` property (inverse of `retrievable`).
+
+`SearchFieldDataType.Collection` is a `staticmethod` monkey-patched onto the generated enum.
+
+## Backward-Compatible Enum Aliases
+
+Monkey-patched at module load in `_patch.py` files:
+```python
+SearchFieldDataType.Int32 = SearchFieldDataType.INT32   # camelCase → UPPER
+OcrSkillLanguage.IS = OcrSkillLanguage.IS_ENUM           # Python keyword collision
+ScoringStatistics.Global = ScoringStatistics.GLOBAL_ENUM
+```
+After regeneration, verify the right-hand-side names still exist in the generated enums. If a new enum collides with a Python keyword, add an alias.
+
+## Polymorphic Delete Pattern
+
+All delete/update operations in `indexes/_operations/_patch.py` accept str or model object:
+```python
+def delete_index(self, index, *, match_condition=MatchConditions.Unconditionally, **kwargs):
+    try:
+        name = index.name   # model object
+        return self._delete_index(name=name, etag=index.e_tag, match_condition=match_condition, **kwargs)
+    except AttributeError:
+        name = index         # string
+        return self._delete_index(name=name, **kwargs)
+```
+New resource types need this same wrapper in both sync and async.
+
+## Removed Model Tombstones
+
+`EntityRecognitionSkill`, `SentimentSkill` → `_RemovedModel` subclasses that `raise ValueError` on instantiation. Keep these for backward compat.
+
+## `_convert_index_response` Helper
+
+`list_indexes(select=...)` returns `SearchIndexResponse` (projection type). `_convert_index_response()` maps it to `SearchIndex`, notably `response.semantic` → `semantic_search`. Shared between sync and async via import.