starred/aryx
1
0
Fork 0
mirror of https://github.com/davidkaya/aryx.git synced 2026-08-06 19:58:43 +02:00
Code Issues 3 Packages Projects Releases 10 Wiki Activity

fix: restore real Copilot handoffs

Browse Source 
Replace the broken forced-tool workaround with a repo-local Copilot
agent adapter that merges runtime handoff instructions and tool
declarations into Copilot sessions and projects Copilot tool requests
back into Agent Framework function-call updates.

Also add regression coverage for the adapter and document the runtime
integration detail in the architecture guide.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
This commit is contained in:
David Kaya
2026-03-27 21:07:58 +01:00
co-authored by Copilot
parent 7247a68f24
commit 9d65e3d209
4 changed files with 610 additions and 104 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ARCHITECTURE.md
+2
View File
@@ -136,6 +136,8 @@ Patterns describe how agents collaborate. The architecture supports:
Their runtime semantics follow the Agent Framework orchestration model: sequential and group chat preserve a visible shared conversation, concurrent aggregates multiple independent responses into one turn, and handoff turns can end once the active agent has responded and is waiting for the next user input.
For Copilot-backed agents, Aryx uses a repo-local adapter around the Copilot SDK session layer so handoff routes still behave like Agent Framework handoffs. This is necessary because the upstream `GitHubCopilotAgent` does not currently project run-time handoff tool declarations into Copilot sessions or surface Copilot tool requests back as `FunctionCallContent` for the workflow runtime.
Patterns are shared application data, not renderer-only configuration. That means the same pattern definition can drive validation, persistence, UI rendering, and sidecar execution.
Patterns now persist an explicit graph-backed topology alongside the flat agent list. Agent nodes carry stable agent ids, ordering, and layout metadata, while system nodes such as user input/output, distributor, collector, and orchestrator make mode-specific flow visible in the saved contract.