Guides for working with coding agents

These guides explain how coding agents behave. They are not the Noteheld manual, and you do not need a Noteheld workspace to use them.

Memory for Claude Code and Codex#

Persistent memory for Claude Code and Codex: your options — instruction files, built-in memory, MCP memory servers, ai-memory and Noteheld, and which one fits which job.

Claude Code#

Guides for Claude Code: memory, sessions, subagents, permissions and hooks — the problem-by-problem guides for Claude Code.

Codex#

Guides for Codex: instructions, sessions and configuration — AGENTS.md loading, config.toml precedence, sessions, approvals, hooks and undo.

OpenCode#

Guides for OpenCode: instructions, sessions and permissions — AGENTS.md loading, session resume, permission rules, undo and plugins.

Copilot CLI#

Guides for GitHub Copilot CLI: instructions, sessions and permissions — AGENTS.md loading, session resume, custom agents, permissions, hooks and skills.

Gemini CLI#

Guides for Gemini CLI: instructions, sessions and memory — GEMINI.md and AGENTS.md loading, session resume, permissions, undo, memory, MCP tools, models, hooks, skills and subagents.

Qwen Code#

Guides for Qwen Code: instructions, sessions and tools — QWEN.md and AGENTS.md loading, permission prompts, session resume, local models, MCP tools, skills and subagents.

Cursor#

Guides for Cursor: rules, sessions, permissions and hooks — AGENTS.md and rule loading, skills, chat history, permissions, MCP tools, hooks and what happened to Memories.

Session history#

Export a Claude Code conversation — the readable copy, the full transcript, and how to find an old session.

Back up coding-agent sessions — a copy of Claude Code, Codex and OpenCode sessions you can restore after cleanup, a reinstall or a new machine.

Delete session history — remove sessions on purpose, including what the vendor keeps.

Instruction files#

One AGENTS.md for every coding agent — native reads, imports, symlinks and copies, and which way to share one file across tools.

How long can AGENTS.md or CLAUDE.md be? — the documented limit and the failure mode for each agent.

Do AGENTS.md files actually help? — what the 2026 controlled benchmark found, and what to keep in a file.

Context compaction#

Context compaction: what your coding agent drops — what survives a summary, what gets lost, and how to keep working state.