Pass the call_id to ApplyPatchTool's editor
Maintainers usually reply within 1 day
Nobody has claimed this yet.
Assessment
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Newbie friendliness
- 70/100
- Issue type
- Feature
- Clarity
- Clearly specified
- Activity status
- Active
- Tech stack
- python
- Domain
- backend-api-design
Research direction
Start with ApplyPatchOperation in src/agents/editor.py and follow ApplyPatchAction.execute in src/agents/run_internal/tool_actions.py, including its approval and resume path. Pass the call ID through to every operation created for that tool call, preserving the existing default for editors that do not provide one. Done when an editor can read the same call ID for each operation, including after approval and resume.
Written by the indexing model from the issue text.
Description
Please read this first
- Have you read the docs? Yes.
- Have you searched for related issues? Yes. #1849 added
tool_call_idto function tool hooks and left built-in
tools for later, asking for a minimal reproduction for a specific built-in local tool with a practical consequence.
This is one forApplyPatchTool, and the change it needs doesn't touch the hook context.
Describe the feature
An ApplyPatchEditor can't tell which apply_patch call it is applying. Its methods get an ApplyPatchOperation
with type, path, diff, ctx_wrapper and move_to (src/agents/editor.py:15-23). ApplyPatchAction.execute
has the id (call_id = extract_apply_patch_call_id(...), src/agents/run_internal/tool_actions.py:928) and calls the
editor without it (:1017-1023).
needs_approval gets (ctx, operation, call_id), so today an editor that needs the id has to remember the
operation object in needs_approval and look it up by identity in the editor. That fails on the approval path. After
RunState.approve() and a resumed run, needs_approval isn't called again (:948-962), so the editor gets an
operation it has never seen and can't learn which approved call it belongs to:
from agents import Agent, ApplyPatchTool, Runner
from agents.editor import ApplyPatchOperation
class AuditEditor:
def update_file(self, op: ApplyPatchOperation):
print(getattr(op, "call_id", None)) # None: the operation doesn't say which call it is
return "done"
create_file = delete_file = update_file
agent = Agent(name="a", tools=[ApplyPatchTool(editor=AuditEditor(), needs_approval=True)])
result = Runner.run_sync(agent, "Fix the typo in README.md") # with a model that calls apply_patch
state = result.to_state()
for item in result.interruptions:
state.approve(item)
Runner.run_sync(agent, state) # the editor runs; nothing tells it which approved call_id this write is
So an editor can't write an audit record per call, check that the write it is about to make is the one a person
approved, or report progress per call in a UI.
Proposed shape: a call_id: str | None = None field, last in ApplyPatchOperation, set in
ApplyPatchAction.execute (or in coerce_apply_patch_operations, which receives the tool call). A call with several
operations gives each the same call_id. It's additive: existing editors ignore it, and positional construction
doesn't change.
Context: I maintain Tracekit (https://github.com/Cygnux-Labs/Tracekit), which records and gates agent tool calls; this would let it hold apply_patch calls for a person's approval (today it has to refuse an ask for them) and record each write under the call the model made.
- Dominant language
- Python
- Stars
- 30.2k
- Forks
- 4.9k
- Avg merge
- 12h 50m
- Merged PRs (30d)
- 162
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from openai/openai-agents-python
-
fix(typing): explicitly specialize ToolContext[Any] in ToolFunctionWithToolContextMay be free again @AbarnaaSree claimed this 9 days ago, and no pull request is open. Open
Difficulty 1/5 Under an hour Newbie friendliness 85/100
openai/openai-agents-python#5387 · 4 comments ·
Maintainers usually reply within 1 day
-
function_tool on a callable object uses an inherited or generated class docstring as the tool descriptionPossibly taken @Narendra02053 claimed this today. Open
Difficulty 3/5 1-2 days Newbie friendliness 58/100
openai/openai-agents-python#5443 · 2 comments ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 44/100
openai/openai-agents-python#5416 ·
Maintainers usually reply within 1 day
-
Difficulty 4/5 3-5 days Newbie friendliness 53/100
openai/openai-agents-python#5413 · 6 comments ·
Maintainers usually reply within 1 day
-
enhancement
Difficulty 4/5 3-5 days Newbie friendliness 45/100
openai/openai-agents-python#5412 · 1 comment ·
Maintainers usually reply within 1 day
All issues in openai/openai-agents-python
Similar issues
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
btclib-org/btclib-wallet#587 · 1 comment ·
Maintainers usually reply within 1 day
-
documentation
Difficulty 2/5 1-3 hours Newbie friendliness 68/100
Judgment-Pack/judgment-pack-spec#275 ·
Maintainers usually reply within 1 day
-
ai-generated
Difficulty 2/5 1-3 hours Newbie friendliness 77/100
Maintainers usually reply within 1 day
-
documentation good first issue priority: low size: XS
Difficulty 1/5 1-3 hours Newbie friendliness 92/100
yeongseon/azure-functions-doctor-python#651 ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 65/100
Changaco/python-libarchive-c#146 ·