[pacocartones]

Summary

• A null entry in tools (or in input/messages when the session fingerprint is computed) made generateSessionId and the OpenAI→Claude tool conversion throw a TypeError instead of ignoring the entry. Such entries are now skipped, as an entry without a name already is. A request without null entries behaves as before.
• Root cause: tools and input are z.array(z.unknown()) in the request schema, so a null entry reaches these readers. generateSessionId read t.name on every tools entry and msg.role on every messages/input entry, and openaiToClaudeRequest read tool.function, all without a guard. messages: [null] already answers 400 since fix(chat): reject null/non-object entries in messages[] (#12643) #13755.

Related Issues

• None found: no open or closed pull request or issue covers this (checked today by the touched paths and function names).

Target Branch

• Base is the active release/v3.8.x branch (the highest open release/v*, here release/v3.8.52), not main

Proof Of Fix (required)

• TDD — a test that fails before the fix and passes after it.
  • Test file: tests/unit/null-entries-in-tools-and-input.test.ts (new, 3 cases: a null entry in tools and in input/messages for the session fingerprint, and a null entry in tools for the OpenAI→Claude conversion, which keeps the valid tool).
  • Failing output before the fix (base release/v3.8.52 with only sessionManager.ts and openai-to-claude.ts reverted):

    ✖ generateSessionId ignores a null entry in tools
      TypeError: Cannot read properties of null (reading 'name')
    ✖ generateSessionId ignores a null entry in input and in messages
      TypeError: Cannot read properties of null (reading 'role')
    ✖ OpenAI to Claude drops a null entry in tools and keeps the valid tool
      TypeError: Cannot read properties of null (reading 'function')
    ℹ tests 3
    ℹ pass 0
    ℹ fail 3
    

  • With the fix: 3 of 3 pass.
• Real-environment test: not done; I did not run a request through the server, so I did not measure which HTTP status the TypeError becomes there.

Scope

• Only the files the failing test proves need changing are touched: open-sse/services/sessionManager.ts, open-sse/translator/request/openai-to-claude.ts, the new test and a one-line changelog fragment.
• No test was weakened or deleted.

Validation

• Change type: request handling (session fingerprint and OpenAI→Claude translation)
• Focused tests: every tests/unit/openai-to-claude* and session* test file, run together with the new file: 739 of 739 pass.
• prettier --check on the new test and the changelog fragment, check:ai-attribution --range and check:changelog-integrity pass. eslint on openai-to-claude.ts reports one unused toolNameMap; it is on the base and this change does not touch it.

Reviewer Notes

• Changelog fragment: changelog.d/fixes/PENDING-null-entries-in-tools-and-input.md, one line, left for you to rename.