Repository navigation
Python: Parse fenced JSON from the group chat orchestrator agent - #8850
Merged
Eduard van Valkenburg (eavanvalkenburg) merged 2 commits intoOct 2, 2026
Conversation
When a provider ignores response_format, the orchestrator agent can return its AgentOrchestrationOutput as JSON wrapped in a Markdown code fence. The parser rejected it, so terminate=true was never applied and the group chat ran until max_rounds. _parse_agent_output now also tries the body of the last fenced block as a parse candidate, after the raw text, so existing behavior is unchanged. Fixes microsoft#8814 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shivam Sharma (ShivamSharma43)
temporarily deployed
to
github-app-auth
September 29, 2026 15:12 — with 
GitHub Actions
Inactive
GitHub Actions
Inactive
Shivam Sharma (ShivamSharma43)
temporarily deployed
to
github-app-auth
September 29, 2026 15:12 — with
GitHub Actions
Inactive
GitHub Actions
Inactive
agent-framework-automation
Bot
added
the
python
Member
|
/review |
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
MAF Automated Review — Iteration 1
Result: Findings reported
Scope: full PR (1 commit(s)): 780d5b320174
Model: gpt-5.6-sol
Overview
The change preserves structured and raw JSON precedence, validates every extracted candidate through the existing Pydantic contract, and adds end-to-end coverage for common fenced responses. The fallback still fails for MIME-style fence tags, and its backtracking regex can synchronously stall the event loop on malformed model output. Both risks are confined to text-fallback responses but can prevent the manager decision from being applied.
Reviewed the supplied pull-request change set across correctness, security/reliability, architecture, and failure behavior.
2 verified findings remained after source verification (2 medium) across 1 file. Details are attached to the affected lines below.
Affected areas: python/packages/orchestrations/agent_framework_orchestrations/_group_chat.py
…ntic Replace the group chat fence regex with extract_markdown_fence_bodies in _orchestrator_helpers.py, which both group chat and Magentic's _extract_json now use. - A closing fence must be at least as long as the opening one and end its line, so backticks inside a JSON string value (for example a fenced code block in final_message) no longer end the block, and an outer fence may use four or more backticks. - The whole info string is consumed, so MIME tags such as application/json parse. - The scan never backtracks, so an unterminated fence followed by whitespace no longer stalls the event loop. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shivam Sharma (ShivamSharma43)
requested a review
from sophia-ramsey
as a code owner
October 1, 2026 18:54
Shivam Sharma (ShivamSharma43)
temporarily deployed
to
github-app-auth
October 1, 2026 18:54 — with
GitHub Actions
Inactive
GitHub Actions
Inactive
Eduard van Valkenburg (eavanvalkenburg)
approved these changes
Oct 2, 2026
This branch was previously deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motivation & Context
AgentBasedGroupChatOrchestratorasks the orchestrator agent for anAgentOrchestrationOutputviaresponse_format. Providers that don't honorresponse_format(for example Claude on Bedrock through an OpenAI-compatible Chat Completions endpoint) return the JSON as plain text wrapped in a Markdown code fence._parse_agent_outputcan't decode fenced JSON, so the orchestration decision, includingterminate: true, is never applied and the group chat runs untilmax_rounds.Description & Review Guide
_parse_agent_outputnow also tries the body of the last Markdown code fence (```json, a bare```, or any language tag) as a parse candidate.ValueError.response_formatnow honor the manager's decisions instead of degrading tomax_rounds. No public API changes._parse_last_json_object).This covers suggested fix (1) from the issue. Fixes (2), exposing
retry_attemptsonGroupChatBuilder, and (3), documentation, are left for separate changes.Related Issue
Fixes #8814
Contribution Checklist
breaking changelabel (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.🤖 Generated with Claude Code