feat: make v2 close_agent interrupt only#26612
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 99d019708d
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| .await; | ||
| return Err(collab_agent_error(agent_id, err)); | ||
| } | ||
| Ok(_) | Err(CodexErr::ThreadNotFound(_)) | Err(CodexErr::InternalAgentDied) => Ok(()), |
There was a problem hiding this comment.
When the target is an open-but-unloaded V2 task name, get_status() above returns NotFound and this arm still returns Ok(()); the tool output is successful, but the emitted CollabCloseEnd carries NotFound, which app-server maps to CollabAgentToolCallStatus::Failed in codex-rs/app-server-protocol/src/protocol/event_mapping.rs. In app-server clients this makes the same successful close/interrupt appear as a failed tool call, so use a non-failure event status or adjust the event mapping for this accepted unloaded-agent case.
Useful? React with 👍 / 👎.
| ToolSpec::Function(ResponsesApiTool { | ||
| name: "close_agent".to_string(), | ||
| description: "Close an agent and any open descendants when they are no longer needed, and return the target agent's previous status before shutdown was requested. Idle agents do not count toward the concurrency limit.".to_string(), | ||
| description: "Interrupt an agent's current turn, if any, and return its previous status. The agent remains available for messages and follow-up tasks.".to_string(), |
There was a problem hiding this comment.
This changes V2 close_agent to leave the agent available after only interrupting the current turn, but downstream rendering still treats every completed CloseAgent item as a close: for example codex-rs/tui/src/multi_agents.rs renders the title as Closed. In TUI/app-server-backed UIs, a user who interrupts a running V2 subagent will be told it was closed even though it remains resident and can still receive follow-up tasks, so the event/display semantics need to be updated alongside the new interrupt-only behavior.
Useful? React with 👍 / 👎.
| .services | ||
| .agent_control | ||
| .subscribe_status(agent_id) | ||
| .interrupt_agent(agent_id) |
There was a problem hiding this comment.
For a target that is actively running, interrupt_agent() only submits an Op::Interrupt to that thread's submission queue and returns the submission id; it does not wait for abort_all_tasks to run or for a TurnAborted status update. Because this handler immediately emits CollabCloseEnd and returns success, callers and UIs can observe close_agent as completed while the subagent is still executing tools/model work, which breaks the new interrupt semantics whenever the parent proceeds assuming the child has stopped.
Useful? React with 👍 / 👎.
jif-oai
force-pushed
the
jif/agent-residency-runtime
branch
from
2f32ac2 to
6390bc7
Compare
jif-oai
force-pushed
the
jif/v2-agent-tool-semantics
branch
2 times, most recently
from
af46ed6 to
79bec9b
Compare
jif-oai
force-pushed
the
jif/agent-residency-runtime
branch
from
3287fc0 to
cc46838
Compare
jif-oai
force-pushed
the
jif/v2-agent-tool-semantics
branch
from
79bec9b to
d1e98a1
Compare
jif-oai
changed the title
jif-oai
force-pushed
the
jif/agent-residency-runtime
branch
from
cc46838 to
0ae3083
Compare
jif-oai
force-pushed
the
jif/v2-agent-tool-semantics
branch
from
d1e98a1 to
83bbc0d
Compare
## Summary Mechanically splits `AgentControl` into focused modules so later agent runtime changes are easier to review. The shared lookup, messaging, and completion logic remains in `control.rs`, while spawn-specific code and V1 legacy close/resume behavior move into dedicated files. ## Changes - Extract spawn-agent code into `agent/control/spawn.rs`. - Extract V1-only legacy close/resume behavior into `agent/control/legacy.rs`. - Keep shared control-plane behavior in `agent/control.rs`. - Preserve existing behavior; this PR is intended to be mechanical. ## Stack 1. This PR - Mechanical `AgentControl` split: extracts spawn and V1 legacy code without behavior changes. 2. #26614 - Execution slot accounting: separates logical agents from active execution slots. 3. #26611 - Residency and reload runtime: adds resident-agent LRU, eviction/reload, durable lookup, and V2 delivery through reload. 4. #26612 - V2 tool semantics: narrows `close_agent` to interrupt-only and updates V2 tool coverage.
1 participant
Summary
Updates V2 agent tool semantics so
close_agentonly interrupts the target agent instead of recursively closing spawned descendants. This keeps V2 task paths and residency behavior intact while making close behavior explicit and less destructive.Changes
close_agentto interrupt only the selected agent.list_agentsas resident-only for V2.Stack
AgentControlsplit: extracts spawn and V1 legacy code without behavior changes.close_agentto interrupt-only and updates V2 tool coverage.