fix(tools): accept zero-argument OpenAI tool calls and reject non-object args

[Agnik47]

Summary

• treat arguments: "" as "no arguments" so the three required: [] tools are callable
• reject non-object JSON at the gate instead of letting it throw out of executeToolCall
• eight regression tests, six of which fail against main

Why

getProfile, documentList and memoryForget all declare required: [] in memoryToolSchemas, so a model may call them with no arguments. The OpenAI API serialises that as arguments: "", and parseToolArguments (added in #1594, cherry-picking #1488) hands the empty string straight to JSON.parse:

{"success":false,"error":"Invalid JSON arguments for getProfile"}

So those three tools never work in their documented no-argument form — the model gets a hard error back and no call is made.

The same gate also lets non-object JSON through, which is the case #1488 was meant to close. "null" parses cleanly, then rejects in the destructuring parameter of every tool function:

TypeError: Cannot destructure property 'containerTag' of 'object null' as it is null.

executeToolCall has no catch, so that rejection escapes and fails the whole request — exactly the throw the gate exists to contain. "5" and "\"text\"" are quieter but worse: they destructure to undefined and call the API with no container tag.

Change

packages/tools/src/openai/tools.ts — blank arguments parse as {}, and the parsed value must be a non-null, non-array object. Malformed JSON still returns the tool error #1488 introduced.

Validation

• vitest run src/ in packages/tools — 42 passed; src/tools.test.ts still fails on a missing SUPERMEMORY_API_KEY, which is pre-existing on main and unrelated
• reverted openai/tools.ts to the main version and re-ran: 6 of the 8 new tests fail, one with the raw TypeError above. The two guard tests — malformed JSON, and an ordinary well-formed call — pass both before and after, so fix(tools): handle malformed OpenAI tool arguments #1488's behaviour is pinned rather than changed
• tsc --noEmit on packages/tools: identical to the main baseline (the three pre-existing openai/tools.ts errors at 329/336/369 remain, tracked in @supermemory/tools fails check-types: 31 errors in test/, including an import of a file that no longer exists #1545; no new ones)
• biome check clean on openai/tools.ts

[yesprasad]

DeepGraph consequence review
DeepGraph traced the changed parseToolArguments helper in:
packages/tools/src/openai/tools.ts:555
The change reaches createToolCallExecutor and downstream package and test surfaces, including:

• packages/tools/src/openai/index.ts
• packages/tools/src/index.ts
• packages/tools/test-supermemory.ts
• packages/tools/test/chatapp/app/api/openai-chat/route.ts
• packages/tools/test/openai-responses-test.ts

The workspace analysis completed successfully:

• 11 projects discovered
• Bun + Turborepo workspace
• 49 workspace imports resolved
• 0 unresolved workspace imports
• Maximum dependency depth: 3

The existing tests already cover empty, whitespace-only, malformed, primitive, array, null, and valid object arguments.
The remaining validation question is whether the intended parser behavior is preserved across every downstream executor and runtime surface identified above—particularly the chat route, Supermemory example, package exports, and response tests.
This is an evidence-based consequence report, not a claim that the implementation is incorrect.