LaunchKit · 2026
Back to Skills

agent-context-system

A persistent local-only memory system for AI coding agents.

0
566 downloads
by @andreagriffiths11

Setup & Installation

openclaw skills install @andreagriffiths11/agent-context-system

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install agent-context-system

Version History

v1.3.2Feb 13, 2026 - Version note by skill developer:

Fix metadata mismatch: root SKILL.md now declares requires.bins: [bash] matching openclaw/SKILL.md. Resolves scanner inconsistency flag.

What This Skill Does

A two-file convention for giving AI coding agents persistent memory across sessions. AGENTS.md holds committed project knowledge shared with the team; .agents.local.md is a gitignored personal scratchpad that grows session by session. No plugins, servers, or background processes required.

Unlike tool-specific memory features, the two-file convention works across Claude Code, Cursor, Copilot, and Windsurf in the same project without any additional tooling.

When to use it

  • Preserving project context so agents don't cold-start each session
  • Sharing architecture patterns and gotchas with teammates via git
  • Logging dead ends so agents don't retry approaches that already failed
  • Coordinating parallel subagents that all share the same project knowledge
  • Gradually promoting recurring session learnings into permanent project docs

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User starts a new coding session on an existing project

AGENT
  1. 1Agent reads AGENTS.md for committed project patterns, commands, and boundaries
  2. 2Agent reads .agents.local.md for personal preferences, past dead ends, and session logs
  3. 3Agent completes the task using accumulated context from both files
  4. 4Agent proposes a session log entry to the user before appending to .agents.local.md
  5. 5Agent writes the approved entry and flags any patterns that have recurred across 3+ sessions as ready to promote
OUTPUT

Session knowledge is logged locally; stable patterns are flagged for the user to promote into AGENTS.md