fix: normalize null tool input from Anthropic adapter

[KrunchMuffin]

#265

Summary

When Claude occasionally produces a tool_use content block with no input_json_delta events (or with "null" as the partial JSON), the Anthropic adapter's content_block_stop handler passes null as the parsed input instead of defaulting to {}.

This causes executeToolCalls to fail Zod schema validation (since null isn't an object). The error result is correctly captured, but the agent loop stalls after the failed tool execution — the model either doesn't produce a follow-up text response, or the follow-up text isn't emitted through the stream.

The end result for users: the assistant says "Let me search for that..." then silence. No error, no follow-up.

Root Cause

In @tanstack/ai-anthropic — adapters/text.ts

The processAnthropicStream method, inside the content_block_stop handler:

parsedInput = existing.input ? JSON.parse(existing.input) : {}

When existing.input is the string "null" (from the model streaming partial_json: "null"), JSON.parse("null") returns JavaScript null — which passes the truthy check but isn't {}. The resulting TOOL_CALL_END event has input: null.

In @tanstack/ai — tools/tool-calls.ts

Downstream in executeToolCalls:

1. toolCall.function.arguments is "null" (set by completeToolCall via JSON.stringify(null))
2. argsStr = "null".trim() || "{}" → "null" (truthy, so no fallback)
3. JSON.parse("null") → null
4. parseWithStandardSchema(tool.inputSchema, null) → Zod throws (null is not an object)
5. Error result pushed, fed back to model
6. Agent loop stalls — no follow-up text produced

Changes

Anthropic adapter (content_block_stop): Normalize parsed input to {} when JSON.parse produces null or any non-object value:

const parsed = existing.input ? JSON.parse(existing.input) : {}
parsedInput = parsed && typeof parsed === 'object' ? parsed : {}

Anthropic adapter (formatMessages): Same normalization when reconstructing tool calls for conversation history (prevents input: null being sent back to the Anthropic API on subsequent turns).

executeToolCalls + ToolCallManager.executeTools: Normalize "null" argument strings to "{}" before JSON parsing — a defense-in-depth fix for the same issue at the tool execution layer.

How to Reproduce

This is intermittent and depends on model behavior. We've observed it with claude-sonnet-4-5 when the model generates a tool_use block with no meaningful arguments (e.g., after a first tool call that returned zero results). The model seems to "start" a follow-up tool call but doesn't commit to arguments.

Typical pattern:

1. User asks a question that triggers a tool call
2. Model produces text + first tool_use (with valid arguments) + second tool_use (with null/empty arguments)
3. First tool executes successfully, second fails validation
4. Agent loop feeds error back to model
5. No follow-up text is produced — stream ends silently

Example Session Data

{
  "role": "assistant",
  "content": "Let me search for disc mowers near you in Saskatoon:",
  "toolCalls": [
    {
      "id": "toolu_1770336584303_0",
      "type": "function",
      "function": {
        "name": "search_equipment",
        "arguments": "null"
      }
    }
  ]
}

Test Plan

• Verified fix handles JSON.parse("null") → normalizes to {}
• Verified fix handles empty existing.input → defaults to {}
• Verified normal tool calls with valid JSON input are unaffected
• All three fix locations use consistent normalization logic

Summary by CodeRabbit

• Bug Fixes
  • Tool argument parsing now treats empty, whitespace-only, and special literal inputs as empty objects, preventing JSON parse errors during tool execution.
  • Improved validation to ensure tool inputs are always treated as objects, defaulting to an empty object when parsing fails or yields a non-object value.

[coderabbitai]

📝 Walkthrough

Walkthrough

Validated and normalized JSON inputs for tool-handling: the anthropic text adapter now ensures parsed JSON is an object before assignment, and the tool-call handlers trim input, convert the literal "null" to "{}", and parse safely to avoid non-object or empty inputs being used.

Changes

Cohort / File(s)	Summary
Anthropic Text Adapter packages/typescript/ai-anthropic/src/adapters/text.ts	Validate parsed JSON is an object; if parsing yields a non-object or fails, assign {}.
Tool Call Argument Parsing packages/typescript/ai/src/activities/chat/tools/tool-calls.ts	Normalize tool argument strings by trimming, treating empty/whitespace or 'null' as '{}', then parse JSON; ensures only object inputs are used and improves parse error handling.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Poem

🐇 I trimmed the strings and hugged the braces tight,
No more lone "null" in the pale moonlight.
Objects only hop through the gate,
Parsers smile — no more uncertain fate.
🥕✨

🚥 Pre-merge checks | ✅ 3

✅ Passed checks (3 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically describes the main fix: normalizing null tool input from the Anthropic adapter, which is the core issue addressed by this PR.
Description check	✅ Passed	The description is comprehensive and follows the template structure with a clear summary of changes, root cause analysis, and test plan, though the checklist items are not marked as completed.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing touches

• 📝 Generate docstrings

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[nx-cloud]

View your CI Pipeline Execution ↗ for commit f0c28a9

---

☁️ Nx Cloud last updated this comment at 2026-02-06 10:32:20 UTC