タグ：context-management

Persistent memory across sessions: hooks record tool calls and generate semantic summaries, then inject relevant context into new sessions, with history search and a web viewer. The hosted memory service needs sign-in, or use your own Anthropic plan or model key.

Gives agents long-term memory across sessions with a local knowledge graph of entities, relations and observations, stored in a local JSONL file you can query and edit. Official MCP reference implementation.

Makes the agent reply in terse 'caveman' style, compressing only prose while keeping code, commands and errors intact, which cuts output tokens significantly. Comes with skills for short commit messages, code review and memory file compression; supports 30+ agents.

Makes the agent answer in terse 'caveman' style to cut output tokens, compressing only prose while leaving code, commands and errors intact. Includes a local proxy that compresses logs, test output and other input. Works with Claude Code, Codex and 30+ other agents.

Rust CLI proxy that filters and compresses command output before it reaches the agent's context, covering git, tests, grep, ls and 100+ other commands and cutting bash output by up to about 90%. Hooks automatically rewrite commands run by Claude Code, Codex and others.

Prebuilds a local knowledge graph of your codebase (symbols, call chains, dependencies) so coding agents get relevant source and change impact in one call, kept in sync as files change. Fully local, supports dozens of languages; run codegraph init in each project first.

Upstash's official Context7 CLI (ctx7): search libraries and pull up-to-date docs from the terminal, or set up the Context7 MCP and rules for Claude Code, Cursor and other agents in one step. Docs lookup needs no login; signing in raises rate limits.

Indexes a codebase into a persistent knowledge graph with tree-sitter so agents can query call chains, impact, architecture and dead code across 162 languages. A single native binary that runs fully locally with no API key; the installer configures detected clients automatically.

Context-driven development plugin: first captures product vision, tech stack and workflow docs, then generates a spec and phased plan for each feature, executes with TDD and supports per-task rollback. Ported from Google's Conductor for Gemini CLI.

Connects Upstash's hosted Context7 remote MCP to fetch current, version-specific docs and code examples for libraries on demand, reducing outdated or invented APIs. No local Node needed; works anonymously, and an API key raises limits. Maintained by Upstash.

Audits whether CLAUDE.md matches the current codebase and scores its quality, and /revise-claude-md writes lessons from the current session back into project memory. Official Anthropic plugin.

Gives coding agents IDE-like symbol-level code search, editing and refactoring via language servers (LSP) across 40+ languages, with cross-session memory. An optional paid JetBrains plugin backend adds breakpoint debugging. Install with uv first.

Runs commands and scripts in a sandbox, indexes large tool outputs for full-text search and returns only results to the context window to save tokens; after compaction it can retrieve edits and task state. Supports Claude Code, Codex and more. Needs Node.js 22.5+.

Processes large tool outputs in a sandbox and returns only the results to context; logs session events in SQLite with FTS5 so work continues after context compaction. Supports Claude Code, Codex and a dozen other clients. ELv2 licensed.

Packs a whole repository into a single AI-friendly file (XML, Markdown, JSON or plain text) with per-file token counts, automatically excludes files containing secrets, and can compress code with Tree-sitter or pack remote GitHub repos directly.

Terminal coding agent from Charm: switch models mid-session while keeping context, manage multiple sessions per project, pull code context via LSP, and extend with MCP and any OpenAI- or Anthropic-compatible API.

Status line plugin for Claude Code showing the model, Git branch, context usage and rate limits live below the input, optionally with running tools, active subagents and todo progress.

Graph-based task tracking and memory for coding agents: a dependency graph replaces Markdown plans, agents claim ready work with bd ready and keep context across sessions. The plugin injects workflow instructions through hooks. Requires the bd CLI.

Graph-based task tracker CLI (bd) for coding agents: replaces Markdown plans with a dependency-aware task graph that gives long tasks persistent memory. Stored in Dolt and syncable across machines; bd init hooks it into Claude Code and Codex automatically.

Manus-style file-based planning: writes the plan, findings and progress to three Markdown files and injects them into context each turn via hooks, so the agent resumes from the current phase after /clear, compaction or a crash. Supports Claude Code, Codex and many other agents.

Keeps the plan, research findings and progress in three files (task_plan.md, findings.md, progress.md) and injects them every turn via hooks, so work resumes after /clear, context compaction or an interrupted session. A Chinese-language variant is also available.

Context and harness engineering skills covering context compression and optimization, multi-agent architectures, memory systems, tool design and LLM-as-judge evaluation, activated automatically by task. Aimed at building production-grade agent systems.

Context and harness engineering skills: context fundamentals, degradation diagnosis and compression, multi-agent architectures, memory systems, tool design, LLM-as-judge evaluation and self-improvement loops. Focused on principles and design patterns, not tied to any agent platform.

Customizes the Claude Code status line to show model, Git branch, token and context usage, session and weekly limits at the bottom of the terminal. Configured through an interactive TUI with Powerline styling. Claude Code only.