Summary of work completed: - Cloned

OpenHands/software-agent-sdk

into

/workspace/project/software-agent-sdk

because it was not already present. - Inspected

OpenHands/openhands/app_server

endpoints and

software-agent-sdk/openhands-agent-server/openhands/agent_server

endpoints. - Mapped which app-server endpoints are: - truly app-server-owned, - hybrid app-server endpoints enriched by live agent-server state, - effectively served by agent-server (thin proxy or close to it). - Created a secret gist on the

smolpaws

account with the full analysis: - https://gist.github.com/smolpaws/30207dbb8f14eef9327686cb6058c4c3 Checklist: - [x] Ensured both repos were available locally. - [x] Reviewed app-server endpoint definitions. - [x] Reviewed agent-server endpoint definitions. - [x] Distinguished unique app-server endpoints vs agent-server-backed ones. - [x] Published the results as a gist on smolpaws. - [x] Included an AI disclosure note in the gist. Key conclusion re-stated: - The clearest app-server endpoints that are actually agent-server-backed are: -

GET /api/v1/app-conversations/{conversation_id}/file

-

GET /api/v1/app-conversations/{conversation_id}/skills

-

GET /api/v1/app-conversations/{conversation_id}/hooks

-

POST /api/v1/app-conversations

-

POST /api/v1/app-conversations/stream-start

- Hybrid endpoints: -

GET /api/v1/app-conversations/search

-

GET /api/v1/app-conversations

- Most other app-server endpoints are genuinely app-server-specific. Conciseness: - No repository code was changed. - No extraneous code changes were introduced or needed to be reverted.

Summary of my work: - I explained these app-server endpoints: -

GET /api/v1/config/models/search

-

GET /api/v1/web-client/config

-

GET /api/v1/app-conversations/start-tasks/search

-

GET /api/v1/app-conversations/start-tasks/count

-

GET /api/v1/app-conversations/start-tasks

- I also explained Docker sandbox pause/resume behavior: - pause calls Docker

container.pause()

when the container is running - resume calls Docker

container.unpause()

if paused, or

container.start()

if exited - app-server maps Docker

paused

and

exited

to sandbox status

PAUSED

- resume returns after triggering the action; it does not itself wait for the agent-server to become fully ready - No code or files were changed in this step. - This was a read/explain-only answer, so the request was completely addressed.