Hi,
Based on the behavior described, this appears to be a limitation of how Copilot Studio generative orchestration handles large tool outputs rather than an issue with Azure DevOps or Power Automate itself.
What is happening?
Your Power Automate flow successfully retrieves the Azure DevOps work items and returns the complete JSON payload. However, when the payload becomes large, Copilot Studio appears to summarize or replace parts of the tool output before passing it to the next prompt or tool.
The placeholder you are seeing:
```text
[truncated workitemjson from RetrieveADOWorkItem]
```
strongly suggests that the orchestration layer is reducing the payload to fit within the model's available context window rather than passing the raw JSON unchanged.
Is there a documented size limit?
Microsoft documents a **5 MB connector payload limit**, but this should not be interpreted as a guarantee that a 5 MB response can be passed unchanged between tools within a generative orchestration flow. The connector payload limit and the LLM context limit are two different constraints. The orchestration engine may still summarize, truncate, or compress tool outputs when they become too large for the model context.
Source:
- Copilot Studio Quotas and Limits
- Prompt Performance and Execution Guidance
Can Copilot Studio automatically summarize tool outputs?
Based on the documented behavior of generative orchestration, yes.
Generative orchestration uses an LLM-based planner that selects tools, manages data flow between them, and builds execution plans dynamically. Large outputs can be condensed before being used by subsequent prompts or tools in order to remain within model limits.
Unfortunately, there is currently no documented setting that forces Copilot Studio to always pass the complete raw tool output to downstream tools.
Recommended architecture
For scenarios involving large structured payloads such as Azure DevOps work items, release-note generation, audit reports, or document creation, the recommended pattern is:
```text
Flow 1
Retrieve Azure DevOps Work Items
↓
Persist full JSON
(SharePoint / Dataverse / Blob Storage)
↓
Return only:
- Document ID
- File URL
- Reference Key
↓
Custom Prompt / Next Tool
↓
Read full content using the reference
↓
Generate Release Notes
↓
Generate DOCX
```
This approach keeps the large payload out of the agent context while preserving access to the complete dataset.
Recommended pattern vs. anti-pattern
✅ Recommended
```text
Tool A
↓
Store full JSON externally
↓
Pass ID/Reference
↓
Tool B reads original data
```
❌ Not recommended
```text
Tool A
↓
Large JSON payload
↓
Prompt
↓
Tool B
```
I believe you are encountering an expected limitation of the generative orchestration runtime rather than a bug.
If preserving the complete Azure DevOps payload is critical, I would recommend persisting the work items externally (SharePoint, Dataverse, Blob Storage, SQL, etc.) and passing only lightweight identifiers between tools. This is the most reliable pattern and avoids context-window limitations, truncation, and unexpected summarization of structured JSON data.
