The apply_diff tool fails with "Missing value for required parameter 'path'" under two conditions

[german-mar]

Problem (one or two sentences)

The apply_diff tool fails with "Missing value for required
parameter 'path'" under two conditions:

1. When the "diff" parameter appears before "path" in the
   parameter list (even with a single diff call).

2. When 2+ apply_diff calls are batched in a single message
   (even with correct parameter order).

---

STEPS TO REPRODUCE

---

TEST RESULTS (4 scenarios)

Test 1: Single diff, path first .......... PASS
Test 2: Single diff, diff first .......... FAIL (Missing path)
Test 3: 2 diffs, path first in both ..... FAIL (Missing path)
Test 4: 2 diffs, diff first in both ..... FAIL (Missing path)

---

WORKAROUND

Context (who is affected and when)

When AI apply a diff

Reproduction steps

• OS: Windows 10
• Shell: cmd.exe
• Java project (Maven)
• Workspace: local filesystem

Bug 1 — Parameter order:

1. Send a single apply_diff where "diff" is the first
   parameter and "path" is second.
2. Expected: diff is applied.
3. Actual: "Missing value for required parameter 'path'".

Bug 2 — Batching:

1. Send 2 apply_diff calls in a single message, both with
   "path" as the first parameter.
2. Expected: both diffs are applied.
3. Actual: both fail with "Missing value for required
   parameter 'path'".

Note: Earlier in the same session, multi-diff messages DID
work. The bug started manifesting mid-session, suggesting
state corruption in the message parser.

Expected result

diff executed

Actual result

Test 1: Single diff, path first .......... PASS Test 2: Single diff, diff first .......... FAIL (Missing path) Test 3: 2 diffs, path first in both ..... FAIL (Missing path) Test 4: 2 diffs, diff first in both ..... FAIL (Missing path)

Variations tried (optional)

No response

App Version

0.0.8162

API Provider (optional)

None

Model Used (optional)

No response

Roo Code Task Links (optional)

No response

Relevant logs or errors (optional)

[roomote-v0]

I have been looking into this issue and have a few questions before I can propose a fix.

Version clarification: You mentioned version 0.0.8162, but the current release is v3.53.0. The tool-call parsing architecture was significantly reworked between those versions -- XML-based tool parsing was removed entirely in favor of native (JSON-based) tool calling. The symptoms you describe (parameter order sensitivity, batching failures) are consistent with the older XML parser behavior, which was inherently order-dependent. Could you confirm:

1. Which version are you actually running? You can check this in the extension panel in VS Code (Extensions sidebar > Roo Code). The latest stable release is v3.53.0.
2. Which API provider and model are you using? (e.g., Anthropic/Claude, OpenAI/GPT-4, Gemini, a local model via Ollama/LM Studio, etc.) This matters because the tool calling mechanism depends on the provider.
3. Can you reproduce this on the latest version (v3.53.0)? The native tool calling system used in the current version parses JSON objects, which are inherently order-independent. If the issue is specific to the older version, upgrading should resolve it.
4. Could you share the relevant log output? In VS Code, open the Output panel (View > Output) and select "Roo Code" from the dropdown. The logs around the time of the error would help identify whether the model is producing malformed tool calls or if there is a parsing issue on our side.

If this issue persists on the latest version, the most likely cause would be either (a) the model itself producing incomplete/malformed JSON for the tool call arguments, or (b) a race condition in the streaming parser when finalization fails and the last partial state is used as the final result. I have identified a potential edge case in NativeToolCallParser where, if JSON finalization fails and the diff parameter appears before path in the streamed JSON, the fallback partial state would have nativeArgs.path as undefined, which would trigger the "Missing value for required parameter path" error.

Please tag @roomote in your response with the answers, and I will be happy to investigate further or implement a fix.