microsoft / microsoft/work-iq

A2A protocol returns no citations in developer telemetry (message/send + message/stream) — CLI/BizChat do; contradicts m365-semantics citations.md

Open
#153 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
PowerShell
Stars
1k
Forks
132
Avg merge
5d 19h
Merged PRs (30d)
6

Description

Summary

The A2A protocol does not return citations in its developer telemetry. For the same agent + prompt where the workiq CLI surfaces citations, an A2A client gets no citation data: no inline [n] markers, no references[], and an empty Answer artifact citation map. This appears to contradict the documented A2A m365 semantics: https://microsoftgraph.visualstudio.com/Graph%20Developer%20Experiences/_git/work-iq-docs?path=/A2A/m365-semantics/citations.md.

Scope — A2A protocol (endpoint https://workiq.svc.cloud.microsoft/a2a/)

Tested both A2A methods with params.configuration.acceptedOutputModes = ["text/plain","application/vnd.ms-workiq.developer-info","application/vnd.ms-workiq-internal.retrieval"], header X-variants: feature.EnableA2AServer:

  • message/send -> no references, no inline [n], no targetLink.
  • message/stream -> artifacts [Answer, Developer Info, Retrieval Info]; Retrieval has 2 execs/~4 hits; Answer citation map = 0 entries; no inline [n]; hits have no webUrl.
Control — workiq CLI (same agent, same prompt) DOES return citations

workiq ask -q "When is my next holiday?" --agent-id <id> --developer --json returns a top-level references array ({ id, targetLink, isCitedInResponse:true }) and inline [1](https://…Holidays_Canada.docx) in the answer. BizChat renders the same source chips. So the citation data exists end-to-end — the A2A protocol surface is the only place it's missing.

Expected

A2A developer telemetry should expose citations per the m365-semantics citations.md — e.g. a references/citation map (with targetLink + isCitedInResponse) on message/send and on the message/stream Answer artifact — at parity with the CLI/BizChat.

Impact

A2A consumers (dev UIs, eval tools) can't measure citation/grounding. The M365 Copilot evals Citations evaluator scores 0 over A2A. Companion: microsoft/m365-copilot-eval#31.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by comparing the A2A message/send and message/stream responses from the documented endpoint with the workiq ask --developer --json control output and m365-semantics/citations.md. Trace how references and the Answer artifact citation map are assembled; done means both A2A methods expose citation data with targetLink and isCitedInResponse, including inline markers where applicable.

Written by the indexing model from the issue text.

Assessment

Domain
api
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
48/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.