Deliver File Sync prompt context to the first workflow task

application/single_app/config.py

@@ -96,7 +96,7 @@
 EXECUTOR_TYPE = 'thread'
 EXECUTOR_MAX_WORKERS = 30
 SESSION_TYPE = 'filesystem'
-VERSION = "0.250.225"
+VERSION = "0.250.226"
 IS_DEVELOPMENT = is_development_env_enabled()
 
 SESSION_COOKIE_SAMESITE = os.getenv('SESSION_COOKIE_SAMESITE', 'Lax')

application/single_app/functions_workflow_runner.py

@@ -246,6 +246,7 @@
 WORKFLOW_CONVERSATION_ACCESS_ERROR = 'Workflow conversation not found or access denied.'
 WORKFLOW_RUN_CANCELLED_MESSAGE = 'Workflow cancellation was requested.'
 WORKFLOW_TASK_CONTEXT_MAX_CHARS = 12000
+WORKFLOW_FILE_SYNC_CONTEXT_MAX_CHARS = 8000
 
 
 class WorkflowRunCancelledError(BaseException):
@@ -6615,7 +6616,9 @@
     if resolved_action.get('target_mode') != DOCUMENT_ACTION_TARGET_MODE_RECENT and not document_ids:
         return {'workflow': workflow, 'citations': [], 'result_count': 0, 'document_count': 0, 'query': None}
 
-    query = str(workflow.get('task_prompt') or '').strip()
+    # Prefer the task's own instructions. task_prompt also carries injected File Sync context and
+    # previous-task output, which would otherwise become part of the search query itself.
+    query = str(workflow.get('task_search_query') or workflow.get('task_prompt') or '').strip()
     if not query:
         return {'workflow': workflow, 'citations': [], 'result_count': 0, 'document_count': 0, 'query': None}
 
@@ -6995,6 +6998,21 @@
     }
 
 
+def _truncate_workflow_file_sync_context(value, max_chars=WORKFLOW_FILE_SYNC_CONTEXT_MAX_CHARS):
+    """Bound the File Sync context block so a large sync cannot dominate the prompt."""
+    normalized = str(value or '').strip()
+    if len(normalized) <= max_chars:
+        return normalized
+
+    head_length = max_chars // 2
+    tail_length = max_chars - head_length
+    return (
+        f'{normalized[:head_length].rstrip()}\n\n'
+        '[File Sync context truncated]\n\n'
+        f'{normalized[-tail_length:].lstrip()}'
+    )
+
+
 def _format_workflow_file_sync_context(file_sync_result):
     if not isinstance(file_sync_result, dict) or not file_sync_result.get('enabled'):
         return ''
@@ -7027,7 +7045,9 @@
         )
     if len(changed_documents) > 50:
         lines.append(f'Additional changed documents omitted from prompt context: {len(changed_documents) - 50}')
-    return '\n'.join(lines)
+    # Truncate here rather than at the injection site so the conversation transcript and the
+    # prompt the model actually receives stay identical.
+    return _truncate_workflow_file_sync_context('\n'.join(lines))
 
 
 def _apply_file_sync_changed_documents_to_action(action_config, changed_document_ids, group_ids, public_workspace_ids):
@@ -7057,6 +7077,12 @@
     file_sync_context = _format_workflow_file_sync_context(file_sync_result)
     if file_sync_context:
         prepared_workflow['task_prompt'] = f"{workflow.get('task_prompt', '')}\n\n{file_sync_context}".strip()
+        # Task-based workflows overwrite task_prompt per task, so the context has to travel on
+        # its own key for _build_workflow_task_execution_workflow() to inject it.
+        prepared_workflow['file_sync_prompt_context'] = file_sync_context
+        # Keep an un-augmented query source for the legacy no-tasks path, whose document search
+        # would otherwise use the whole changed-document manifest as its search query.
+        prepared_workflow.setdefault('task_search_query', str(workflow.get('task_prompt') or '').strip())
 
     config = _get_workflow_file_sync_config(workflow)
     changed_document_ids = list(file_sync_result.get('changed_document_ids') or [])
@@ -9103,17 +9129,26 @@
     return {'type': DOCUMENT_ACTION_TYPE_NONE}
 
 
-def _build_workflow_task_execution_workflow(workflow, task, previous_reply='', include_document_action=False):
+def _build_workflow_task_execution_workflow(
+    workflow,
+    task,
+    previous_reply='',
+    include_document_action=False,
+    include_file_sync_context=False,
+):
     task = task if isinstance(task, dict) else {}
     prepared_workflow = dict(workflow or {})
     task_instructions = str(task.get('instructions') or '').strip()
+    # Captured before any context blocks are appended so document search queries stay scoped to
+    # what this task actually asks for.
+    task_search_query = task_instructions
     file_sync_context = str(workflow.get('file_sync_prompt_context') or '').strip()
     task_document_action = _resolve_workflow_task_document_action(
         workflow,
         task,
         include_document_action=include_document_action,
     )
-    if include_document_action and file_sync_context:
+    if include_file_sync_context and file_sync_context:
         task_instructions = (
             f'{task_instructions}\n\n'
             '[Workflow input context]\n'
@@ -9129,6 +9164,7 @@
         ).strip()
 
     prepared_workflow['task_prompt'] = task_instructions
+    prepared_workflow['task_search_query'] = task_search_query
     prepared_workflow['active_task'] = {
         'id': str(task.get('id') or '').strip(),
         'name': str(task.get('name') or '').strip(),
@@ -9541,6 +9577,7 @@
                     task,
                     previous_reply=previous_reply,
                     include_document_action=task_index == 0,
+                    include_file_sync_context=task_index == 0,
                 )
                 attempt_workflow, runner_audit = _resolve_workflow_task_runner(
                     prepared_workflow,

docs/explanation/fixes/WORKFLOW_FILE_SYNC_PROMPT_CONTEXT_FIX.md

@@ -0,0 +1,206 @@
+# Workflow File Sync Prompt Context Fix
+
+**Fixed in version: 0.250.226**
+**Issue:** [microsoft/simplechat#1285](https://github.com/microsoft/simplechat/issues/1285)
+
+## Issue
+
+File Sync's **prompt context** — the block that tells a workflow *what actually changed* — never
+reached the model in any task-based workflow. Since the workflow builder always creates at least
+one task, that was effectively every workflow.
+
+The failure was silent and actively misleading: the context **was** written into the
+conversation's user message, so the run transcript displayed the changed-document list as though
+the model had received it. In practice the model was given only the raw task instructions and
+typically replied that it had no information about any documents.
+
+### What File Sync is supposed to do
+
+Before a workflow runs, File Sync scans the configured sources and hands the workflow two separate
+things:
+
+1. **Prompt context** — built by `_format_workflow_file_sync_context()`: the sync counts
+   (`Scanned / Created / Updated / Unchanged / Skipped / Failed`) and a numbered list of each new
+   or changed document (`relative_path`, `action`, `document_id`, `source_name`). This is what
+   makes instructions like *"summarize the documents that changed since the last run"* work.
+2. **Document targets** — when **Use changed documents** is enabled and the document action is
+   Analyze, the action is repointed at exactly the changed document ids.
+
+Item 2 worked. Item 1 did not.
+
+### Who was affected
+
+- **Monitor File Sync Changes** workflows — an entire trigger type whose purpose is reacting to
+  what changed — unless they happened to use Analyze with **Use changed documents**.
+- Any workflow using `Search` or `No document action` got nothing at all from File Sync.
+- Any workflow with **Use changed documents** disabled got nothing.
+
+## Root Cause
+
+`_apply_file_sync_context_to_workflow()` appended the context to the **workflow-level**
+`task_prompt`. `_build_workflow_task_execution_workflow()` expected it on a **different key**, then
+overwrote `task_prompt` with the task's own instructions:
+
+```python
+file_sync_context = str(workflow.get('file_sync_prompt_context') or '').strip()  # always ''
+if include_document_action and file_sync_context:                                # never fired
+    task_instructions = f'{task_instructions}\n\n[Workflow input context]\n{file_sync_context}'
+...
+prepared_workflow['task_prompt'] = task_instructions   # discarded the appended context
+```
+
+`file_sync_prompt_context` was read in exactly one place and written nowhere. The producer line was
+added in `79148c84` ("Add stepped workflow builder and task sequences") alongside the consumer:
+
+```diff
+     if file_sync_context:
+         prepared_workflow['task_prompt'] = f"{workflow.get('task_prompt', '')}\n\n{file_sync_context}".strip()
++        prepared_workflow['file_sync_prompt_context'] = file_sync_context
+```
+
+It was later lost in a merge resolution on a long-lived branch (the
+`fix/1031-tabular-row-orchestration-scale` / PR #1145 lineage). The consumer survived; the producer
+did not.
+
+### Why no test caught it
+
+`functional_tests/test_workflow_task_sequence.py` hand-injected `file_sync_prompt_context` directly
+into the workflow dict and then asserted that `[Workflow input context]` appeared in the dispatched
+prompt. It exercised the consumer with a value production never supplied, so it passed while the
+feature was broken.
+
+## Technical Details
+
+### Files Modified
+
+| File | Change |
+|---|---|
+| `application/single_app/functions_workflow_runner.py` | Restored the producer, bounded the context, gave the search query a clean source, made the first-task gate explicit |
+| `application/single_app/config.py` | `VERSION` `0.250.225` → `0.250.226` |
+| `functional_tests/test_workflow_file_sync_prompt_context.py` | **New** producer-to-consumer coverage |
+| `functional_tests/test_workflow_task_sequence.py` | Builds the context through the real producer instead of injecting the key |
+
+### Code Changes
+
+**1. Restored the producer.**
+
+```python
+if file_sync_context:
+    prepared_workflow['task_prompt'] = f"{workflow.get('task_prompt', '')}\n\n{file_sync_context}".strip()
+    prepared_workflow['file_sync_prompt_context'] = file_sync_context
+    prepared_workflow.setdefault('task_search_query', str(workflow.get('task_prompt') or '').strip())
+```
+
+**2. Gave the document search query a clean source.**
+
+This was the blocking discovery. `_prepare_workflow_search_context()` used `workflow['task_prompt']`
+**verbatim as the Azure AI Search query**, and that single `query` value fed all four search call
+sites. Restoring the injection without this change would have turned a Search task's query into 50
+lines of file paths and sync counters.
+
+The prepared workflow now carries `task_search_query`, captured from the task's instructions
+*before* any context blocks are appended:
+
+```python
+task_search_query = task_instructions   # captured before File Sync / previous-output injection
+...
+prepared_workflow['task_search_query'] = task_search_query
+```
+
+```python
+query = str(workflow.get('task_search_query') or workflow.get('task_prompt') or '').strip()
+```
+
+The fallback to `task_prompt` means every caller without the key behaves exactly as before.
+
+This also corrects a related case: after per-task workspace documents shipped in
+[#1284](https://github.com/microsoft/simplechat/pull/1284), a task other than the first can carry a
+Search action, and its query previously included the entire previous-task reply. Search queries are
+now scoped to the task's own instructions.
+
+**3. Made the first-task gate explicit.**
+
+The consumer was gated on `include_document_action`, which used to mean "task 1". After #1284 that
+flag means "legacy record with no task-level document action", so it no longer expressed the
+intent. A dedicated parameter now carries it:
+
+```python
+def _build_workflow_task_execution_workflow(
+    workflow, task, previous_reply='', include_document_action=False, include_file_sync_context=False,
+):
+```
+
+```python
+prepared_workflow = _build_workflow_task_execution_workflow(
+    workflow, task,
+    previous_reply=previous_reply,
+    include_document_action=task_index == 0,
+    include_file_sync_context=task_index == 0,
+)
+```
+
+Later tasks receive the information indirectly, through task one's response, which is already
+chained forward as bounded context.
+
+**4. Bounded the context block.**
+
+`WORKFLOW_FILE_SYNC_CONTEXT_MAX_CHARS = 8000` with head/tail truncation and a
+`[File Sync context truncated]` marker, mirroring the existing treatment of previous-task output.
+It is applied inside `_format_workflow_file_sync_context()` rather than at the injection site, so
+the conversation transcript and the prompt the model receives stay identical — the mismatch between
+those two is exactly what made this bug invisible.
+
+### Testing
+
+`functional_tests/test_workflow_file_sync_prompt_context.py` covers:
+
+- The producer publishes `file_sync_prompt_context`, and attaches nothing when File Sync is off.
+- The first task's prompt carries `[Workflow input context]` and the changed-document list; later
+  tasks do not.
+- The legacy no-tasks path still carries the context on `task_prompt`.
+- Search queries use the task's own instructions, with no File Sync block and no previous-task
+  output, for both the task path and the legacy path.
+- The truncation notice appears past the cap and not below it, including a realistic 50-document
+  sync with deeply nested paths.
+- A run with nothing changed still tells the first task that nothing changed.
+- **Use changed documents** targeting is unaffected.
+
+Both this file and the updated `test_workflow_task_sequence.py` were verified by mutation: removing
+the restored producer line makes 4 of 8 and 1 of 10 tests fail respectively, so the regression
+cannot silently return.
+
+## Validation
+
+### Before
+
+| Scenario | Result |
+|---|---|
+| File Sync workflow, `No document action`, instructions ask what changed | Model has no information about any documents |
+| Conversation transcript | Shows the full changed-document list, implying the model received it |
+| Search task query | The task instructions only, because the context never arrived |
+
+### After
+
+| Scenario | Result |
+|---|---|
+| File Sync workflow, `No document action`, instructions ask what changed | First task's prompt contains the sync counts and changed-document list |
+| Conversation transcript | Matches what the model received |
+| Search task query | The task's own instructions, free of File Sync context and previous-task output |
+| Very large sync | Context bounded at 8000 characters with an explicit truncation notice |
+| Nothing changed | First task is told "No new or changed synced documents were detected." |
+
+### Regression Testing
+
+The full `test_workflow*` suite was run against a clean `git archive` export of the base commit.
+Both baseline and branch produce the same 21 pre-existing failures, and per-file output diffs
+contain only traceback path differences. Route policy tests pass, and the personal and group
+document picker harnesses from #1284 still pass.
+
+## Related
+
+- Issue: [microsoft/simplechat#1285](https://github.com/microsoft/simplechat/issues/1285)
+- Preceding work: [#1284](https://github.com/microsoft/simplechat/pull/1284), which introduced
+  per-task workspace documents and changed the meaning of `include_document_action`
+- Feature: `docs/explanation/features/WORKFLOW_PER_TASK_WORKSPACE_DOCUMENTS.md`
+- Tests: `functional_tests/test_workflow_file_sync_prompt_context.py`,
+  `functional_tests/test_workflow_task_sequence.py`

docs/explanation/fixes/index.md

@@ -30,3 +30,4 @@ category: Version History
 - [Route Authentication Audit Findings Fix](ROUTE_AUTHENTICATION_AUDIT_FINDINGS_FIX.md)
 - [Cosmos Container Throughput Deployer Fix](COSMOS_CONTAINER_THROUGHPUT_DEPLOYER_FIX.md)
 - [Workflow Task Document Picker Fix](WORKFLOW_TASK_DOCUMENT_PICKER_FIX.md)
+- [Workflow File Sync Prompt Context Fix](WORKFLOW_FILE_SYNC_PROMPT_CONTEXT_FIX.md)

docs/explanation/release_notes.md

@@ -2,6 +2,22 @@
 
 For feature-focused and fix-focused drill-downs by version, see [Features by Version](/explanation/features/) and [Fixes by Version](/explanation/fixes/).
 
+### **(v0.250.226)**
+
+#### Bug Fixes
+
+*   **File Sync Now Tells the Workflow What Changed**
+    *   File Sync builds a summary of each run — the scan counts plus every new or changed document — but that summary never reached the model in any workflow that uses tasks, which is every workflow the builder creates.
+    *   The failure was silent and misleading: the summary *was* written into the conversation, so the transcript showed the changed-document list as though the model had received it. In practice the model saw only the raw task instructions and usually replied that it knew nothing about any documents.
+    *   This hit **Monitor File Sync Changes** workflows hardest, along with any workflow using Search or no document action, or with **Use changed documents** turned off. The first task in the sequence now receives the summary, and later tasks get it through the first task's response.
+    *   The summary is also bounded now, with a clear truncation notice, so a very large sync cannot crowd out the actual instructions.
+    *   (Ref: #1285, `functions_workflow_runner.py`, File Sync prompt context)
+
+*   **Document Search Queries Are No Longer Diluted by Injected Context**
+    *   A workflow's document search used the entire task prompt as its search query, including the File Sync summary and the previous task's full response. A search for "find the renewal clause" could end up querying 50 lines of file paths.
+    *   Search queries now use the task's own instructions. Retrieved content and context still reach the model exactly as before — only the query is scoped.
+    *   (Ref: #1285, `functions_workflow_runner.py`, workflow document search)
+
 ### **(v0.250.225)**
 
 #### New Features