What problem it solves
AI coding agents forget everything between sessions, lose context when switching tools, and can't share knowledge across teams. Every new conversation starts from zero — repeated explanations, inconsistent output, wasted tokens.
https://lijma.github.io/agent-skill-fcontext/
Who uses this workflow
- Individual developers switching between Copilot, Claude Code, Cursor, and other AI agents
- Teams needing to share domain knowledge and onboard new members' AI agents
- Projects with binary documents (PDF/DOCX/XLSX) that AI agents can't read natively
- Requirements-driven development workflows
How it works
fcontext is an open-source CLI tool (pip install fcontext) that maintains a .fcontext/ directory with structured context data. It:
- Persists session knowledge across conversations (
_topics/)
- Indexes binary documents to Markdown (
_cache/)
- Tracks requirements with evolution history (
_requirements/)
- Exports/imports domain knowledge as experience packs (
_experiences/)
- Generates agent-native instructions for 6 agents (Copilot, Claude, Cursor, Trae, OpenCode, OpenClaw)
All data is local, offline, no API keys needed.