Your team uses different AI coding agents. Alice prefers Cursor, Bob uses Claude Code, and Charlie swears by Copilot. Each agent has its own config format, so you end up with fragmented, inconsistent instructions — or no shared instructions at all.
{ "instructions": [ { "id": "code-style", "title": "Code Style", "content": "Follow the style guide in docs/CODE_STYLE.md. Use ESLint and Prettier configs without override. Variable names in camelCase, components in PascalCase.", "alwaysApply": true }, { "id": "pr-workflow", "title": "PR Workflow", "content": "All changes go through PRs. Branch naming: feature/*, fix/*, chore/*. Require at least one review. Squash merge to main.", "alwaysApply": true }, { "id": "testing-rules", "title": "Testing Standards", "content": "All new features require tests. Minimum 80% coverage for new code. Use vitest for unit tests, playwright for e2e.", "globs": ["**/*.test.ts", "**/*.spec.ts", "tests/**"], "alwaysApply": false } ]}
Commit both the config file and the generated agent files. This way, team members get instructions immediately without needing to run drevon sync first.
The new team member gets everything automatically when they clone the repo — all agent configs are already generated and committed.If they want to add their personal agent:
drevon add-agent windsurf# Don't commit this — it's personal
The v2 project-mode topic files under .drevon/memory/topics/ work well as shared team context.
Commit topics/ — the log/ segments are episodic and can get noisy:
Path
Team Use
INDEX.md
Generated pointer file — commit it; it’s small and keeps everyone oriented
topics/architecture.md
Shared project overview — update during onboarding
topics/decisions/
ADR records, one file per decision — commit these for transparency
topics/patterns.md
Team code conventions — evolve collaboratively
log/
Consider .gitignore-ing these segments for noisy workspaces
Because each decision is its own file under topics/decisions/<date>-<slug>.md, PR review and
merges stay clean — teammates rarely touch the same file, so there are no merge conflicts on a
single monolithic decisions.md.
Memory is designed for single-agent use. In team settings, be intentional about what you commit:
the topics/ tree works well as shared docs, while the episodic log/ segments can get noisy.
