Four agent frameworks shipped the same fix in one week: context compaction vs approvals
HelpfulCollar1070 · reddit · 2026-10-17
A author tracking 40 agent frameworks/SDKs noticed four projects shipped fixes for the same problem during Oct 12–16:
- Haystack 3.4.0: an experimental compactor that moves old tool results into a store, leaving a reference in context instead of summarizing them away.
- OpenAI Agents SDK 0.24.0: pauses automatic compaction while an approval is pending.
- Google ADK 2.12.0: requireconfirmation for human sign-off before workflow nodes run, plus a callback to filter tool output before it reaches the model.
- OpenClaw 2026.10.4: manual /compact now lets the active run finish rather than aborting and dropping the pending request.
From production agent experience, the author argues the real failure mode for long-running agents isn't model capability—it's forgetting what was done 40 steps ago, or killing its own approval request to free context. His take: compaction is the new retry logic. When picking a framework, ask what it does when context fills mid-task and whether a human approval survives that.
More from coding & agent
- Expo SDK 58 ships a Rust JS transformer: 5,477 files converted in under 3 seconds — Baconbrix · 2026-10-17
- LangChain launches Managed Deep Agents: sandbox proxies make production agents secure by default — LangChain · 2026-10-17
- Agent-Reach CLI hits 98k GitHub stars, gives AI agents free access to the whole web — victor_explore · 2026-10-17
- Claude Opus overnight run turns Christmas present into coal, silently deletes files — johnlindquist · 2026-10-17
- LangChain's Managed Deep Agents keeps credentials out of the sandbox via proxy-injected tokens — LangChain · 2026-10-17
- Eazo generates 6 design directions per idea, then builds the working app — Aiden_Tech_Ai · 2026-10-17