Fix 09c_hitl_streaming: approval pause never fires - #471
Merged
Conversation
kowser-orkes
approved these changes
Aug 12, 2026
ling-senpeng13
force-pushed
the
fix/09c-hitl-approval-pause
branch
from
August 12, 2026 21:16
ef5d794 to
7c67c3e
Compare
The agent's instructions said "Only suggest deleting data if explicitly asked", so the model returned the approval question as its final answer (finishReason: STOP) instead of calling the tool. No WAITING event was emitted, so the example's approval handler never ran and the approval_required=True pause it exists to demonstrate never fired. Reworded as ordered steps that name the tool and leave approval to the gate. Verified 4/4 on openai/gpt-4o-mini and anthropic/claude-sonnet-4-5: check -> restart -> approval pause -> delete. A one-word suggest->delete change is not enough: any prose phrasing states the deletion as a permission, and gpt-4o-mini treats a permission as optional — it restarts, reports success, and skips the tool (0/5 across two prose variants). Numbered imperative steps get it executed.
ling-senpeng13
force-pushed
the
fix/09c-hitl-approval-pause
branch
from
August 14, 2026 17:57
7c67c3e to
eeb95e9
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The agent's instructions said "Only suggest deleting data if explicitly asked", so the model returned the approval question as its final answer (
finishReason: STOP) instead of calling the tool. NoWAITINGevent was emitted, so the example's approval handler never ran and theapproval_required=Truepause it exists to demonstrate never fired (0/6 runs).Reworded as ordered steps that name the tool and leave approval to the gate.
Verified 4/4 —
openai/gpt-4o-mini2/2,anthropic/claude-sonnet-4-52/2:check → restart → approval pause → delete.
Why not a one-word
suggest→deletefix: any prose phrasing states the deletion as a permission, andgpt-4o-minitreats a permission as optional — it restarts, reports success, and skips the tool (0/5 across two prose variants). Numbered imperative steps get it executed.