$ the-wire · showcase
AGENT TOOL SIGNATURES NOW ENFORCED BEFORE EXECUTION
By RepoJournal · Filed · About AI Agents
![openhands-release-bot[bot]](https://avatars.githubusercontent.com/u/225919603?v=4)
13 people shipped this
Saoud Rizwan
@saoudrizwan
2 cited
Sydney Runkle
@sydney-runkle
1 cited
Mason Daugherty
@mdrxy
1 cited
openhands-release-bot[bot]
@openhands-release-bot[bot]
1 cited
Mikołaj Kondratek
@mkondratek
1 cited
Srivatsa03
@Srivatsa03
1 cited
YZJF
@YZJF
1 cited
langchain-oss-model-profiles[bot]
@langchain-oss-model-profiles[bot]
Huzaifa Iftikhar
@HuzaifaChaudary
hieptl
@hieptl
neubig
@neubig
nightcityblade
@nightcityblade
mohammed Hafeez
@MUHAMMEDHAFEEZ
The retry logic that could run your failed function tools twice just got the axe.
LlamaIndex is shipping a critical behavior change: `call_tool` and `acall_tool` no longer blindly retry a `FunctionTool` with keyword arguments after any exception from a positional call. The calling convention is now determined from the callable's signature before invocation, preventing irreversible side effects from executing twice on transient failures [1]. That pairs with a metadata bug where a target key present but set to `None` was clobbering node content instead of falling back [2]. Meanwhile, Cline fixed a checkpoint restore that quietly knocked commits off your branch when HEAD had moved past the checkpoint [3], and OpenAI Codex sign-in now fails loudly instead of dead-ending silently when the callback port is in use [4]. LangChain's Fireworks integration drops stale reasoning blocks that caused 400s after switching models [5], and PII redaction no longer destroys the content-block shape of your messages [6]. OpenHands v1.16.0 landed with Linux desktop installer support and a provider picker [7]. Cline's history cost estimates are now hidden for subscription-billed tasks, killing the phantom $X.XXXX rows [8]. Across the stack, 48 commits and 7 releases tell a story of surgical fixes, not big features.
Action items
- → Upgrade llama_index to get the retry fix before you deploy tools with side effects run-llama/llama_index [immediate]
- → Review Cline checkpoint restore usage after HEAD has moved cline/cline [immediate]
- → Update langchain-fireworks to 1.6.1 to avoid 400 errors after model switches langchain-ai/langchain [plan]
- → Watch PIIMiddleware redaction changes if you store message content as blocks langchain-ai/langchain [monitor]
References
- [1] fix(core): avoid retrying failed function tools ↗ run-llama/llama_index
- [2] fix(core): fall back when metadata replacement target value is None ↗ run-llama/llama_index
- [3] fix(core): refuse checkpoint workspace restore when HEAD moved past the checkpoint (#13626) cline/cline
- [4] fix: make OpenAI Codex (ChatGPT subscription) sign-in fail loudly instead of silently dead-ending (#13537) cline/cline
- [5] fix(fireworks): drop reasoning history blocks (#39973) langchain-ai/langchain
- [6] fix(langchain): preserve content-block shape in PIIMiddleware redaction (#39894) langchain-ai/langchain
- [7] v1.16.0 ↗ OpenHands/OpenHands
- [8] fix: hide history cost estimates for subscription-billed tasks (#13562) cline/cline
