fix(grok): encode session/cancel as a real notification and interrupt steers - #224
Conversation
…anges Grok health checks opened a full ACP session, so background probes could hang on authenticate or boot every configured MCP server. They also never reported login state, and the snapshot locked model changes to a new thread even though session/set_model already works. The check now runs grok --version, grok models, then a single ACP initialize and reads models from _meta.modelState. Failed initialize degrades to warning. grok-build is never sent as a model id. Adapted from pingdotgg#9154 and pingdotgg#8392. Made by Grok 4.6 High in Grok Build via Orca.
Keep the Grok probe PR current with main.
… steers Grok CLI dropped session/cancel because effect-acp encoded notifications as requests with an empty id and headers. Stop then queued the next prompt behind work that never cancelled. Notifications now encode as JSON-RPC with no id. Cancel waits for that write. A mid-turn send cancels the in-flight prompt and continues the same turn. Adapted from pingdotgg#9154. Shared ACP encoding also applies to Cursor and OpenCode; their adapters still use start() and are covered by existing tests. Made by Grok 4.6 High in Grok Build via Orca.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
github-actions
Bot
added
type:provider
Greptile SummaryThis update corrects Grok mid-turn steering so a superseded prompt can flush the replacement prompt’s stored terminal outcome when it drains last. The accompanying tests now cover both prompt-settlement orders. Confidence Score: 5/5Safe to merge. No outstanding findings remain. Reviews (3): Last reviewed commit: "merge main into grok cancel stack" | Re-trigger Greptile |
A successful replacement could be reported as cancelled when the superseded prompt finished local draining last. The last in-flight settlement owned the terminal state. Store the current epoch's result and emit that when the last prompt drains. Unit tests cover both settlement orders. Made by Grok 4.6 High in Grok Build via Orca.
|
This is Leo's agent. Independent review found that #228's terminal-emission finding also applies to this exact head, I ran the 109 focused protocol and adapter tests successfully, then reproduced the omitted case against the shipped
The helper is identical in #224 and #228. This reproduction exercises the real helper; it does not claim a full provider-process interleaving. The existing adapter tests do not cover this final no-emission drain. I requested a scoped repair and committed lifecycle regression from the existing Grok worker, with the fix propagated to #228. That worker is currently weekly-quota blocked, so no fix has executed. Both PRs remain on hold pending the repair and fresh verification. |
github-actions
Bot
added
type:plugin
github-actions
Bot
added
size:L
and removed
type:plugin
1 participant
Problem
Grok CLI dropped
session/cancelbecause effect-acp encoded notifications as JSON-RPC requests withid: ""andheaders. Stop did not stop. The next send queued behind the old prompt.Mid-turn sends also waited behind the in-flight ACP prompt instead of interrupting like Claude and Codex.
Changes
Outbound ACP notifications now encode as JSON-RPC with no
idorheaders. RpcClient Interrupt frames are dropped.AcpSessionRuntime.cancelwaits for the cancel write.promptcan signaldispatchedonce the RPC fiber is registered.Grok mid-turn sends cancel the in-flight prompt, then send the replacement under a prompt epoch and lifecycle semaphore. A failed steer does not discard the live prompt.
Upstream
Adapted from pingdotgg/t3code#9154 session/cancel and steer work. Depends on #214 for
grok-buildhandling and initialize-only runtime.Shared ACP encoding also applies to Cursor and OpenCode. Cursor adapter tests still pass. No Mastra Codex/Kimi/OpenCode Go change.
Scope
This PR is cancel encoding and Grok steer only.
Assigned Grok ports:
Verification
Stacked on #214. Leave both open.
Made by Grok 4.6 High in Grok Build via Orca.