context-compaction

Zero Code Studio 是一个工程化 AI 代码生成平台，基于 Spring Boot、LangChain4j、LangGraph4j，支持工具调用、Workflow 编排、任务计划规划、三层上下文压缩、SSE 过程可视化、定点修改与一键本地部署。

A local GUI workaround for Codex Desktop conversations stuck on automatic context compaction.

Six in-process middlewares for OpenClaw: HITL approvals, prompt-injection guardrails, PII redaction, tool-call budgets, context compaction, and complexity-aware model routing. Zero telemetry, all state local.

Zero context compaction for Claude Code & LLM agents. Persistent memory powered by OS primitives (demand paging, kswapd eviction, mlock pinning). Single SQLite file, MCP-native, multi-agent shared.

Substrate-based context compaction for Claude Code. Inspectable signal weights, local-only, MCP-ready. Alternative to /compact + hook-based capture (claude-mem, etc.).

Verified context compaction for coding agents - codex's engine + mechanical fact carriage + the only handoff test in open source. Drop-in pi extension.

Auto-summarize your AI coding session with Anthropic's 9-section template before context compaction. Cross-IDE: Claude Code / Cursor / CodeBuddy / WorkBuddy / Codex Desktop.

Patch Claude Code to use the FlashHash method for compaction -- 1mtok > 50ktok compaction in <10s with perfect context preservation on the cheapest 1M-context model.

Claude Code plugin: self-contained session state preservation across context compaction (state capture, recovery injection, resume recovery, built-in usage warnings)

ClaimKeep — narrow continuous-memory plugin for Claude Code: harvests an agent's high-signal calibrated claims into a schema-stable brief before compaction and re-injects after. Paper (Zenodo): zenodo.org/records/20819013

Pi extension implementing Codex CLI-compatible remote V1/V2 compaction for OpenAI Codex models.

Codex-derived source release for an opt-in session compact route that preserves durable handoff state across repeated compactions.

Keep Codex long tasks running while you sleep: local compact/context recovery with audited bundles and safe one-relay handoffs.

We measure whether an indirect prompt injection buried earlier in a conversation survives an LLM-driven context compaction step (the mechanism production agent platforms use to summarize long conversation histories and discard the original turns), and whether it retains behavioral force afterwards. Across 994 trials over three open-weight summarize

Enhanced auto-compaction defaults for DeepSeek Harness agent presets

Durable memory across context compaction for the pi coding agent: every compaction writes a greppable Markdown record

Qontext — a quipu-inspired conversational memory

Save Token Usage on Unstructured Document 😎. Let agent read docs, memories, prompts with in ultra-compressed mode through a tiny local model.

Codex plugin for task continuity after context compaction

Maintained CLIProxyAPI fork adding opt-in Codex server-side compaction for Claude Code sessions with persistent opaque-artifact replay.