Configuration and data directories

ScienceX stores application data under ~/.sciencex by default. Project configuration lives in .sciencex/ at the project root. Legacy .claude/ paths are compatibility read sources and are no longer the default write targets.

User-level layout

~/.sciencex/
├── config/          # Providers, desktop settings, and IM adapter config
├── credentials/     # OAuth and other credentials
├── state/           # Window, terminal, scheduled-task, and session-map state
├── data/            # Indexes, traces, downloads, and the Science project registry
├── diagnostics/     # Diagnostic reports
├── runtime/         # Temporary runtime files
└── claude/          # Embedded Claude-compatible runtime
    ├── settings.json
    ├── skills/
    ├── agents/
    ├── projects/
    ├── teams/
    └── tasks/

The claude/ subtree is the compatibility boundary for the embedded runtime. It keeps the expected Claude configuration shape without occupying ~/.claude in the user's home directory.

Project-level layout

<project>/.sciencex/
├── settings.json
├── settings.local.json
├── scheduled_tasks.json
├── skills/
├── agents/
├── rules/
├── output-styles/
└── worktrees/

settings.local.json, scheduled_tasks.json, worktrees/, and SQLite temporary files should normally stay out of version control. Shared settings.json, skills, agents, and rules may be committed when appropriate.

Environment variables

Variable	Purpose
SCIENCEX_HOME	Overrides the user-level ScienceX root; defaults to ~/.sciencex
CLAUDE_CONFIG_DIR	Overrides only the embedded Claude-compatible runtime; legacy launchers remain supported
SCIENCEX_LEGACY_CONFIG_DIR	Selects the legacy source for one-time migration; normally unset

Selecting a custom data directory in the desktop app is equivalent to setting SCIENCEX_HOME; the compatible runtime is placed in its claude/ subtree automatically.

Migrating from .claude

On first use of the new layout, ScienceX copies known legacy data with these safeguards:

1. A legacy entry is copied only when its new target is missing; existing .sciencex content always wins.
2. Copies use a temporary path and atomic rename, so a failure does not publish a partial target.
3. Symbolic links are not followed, and unknown credential files are not copied.
4. Existing .claude content is never deleted, renamed, or modified.
5. Project compatibility is per category: a .claude file or directory is read only when the corresponding .sciencex entry is absent, and subsequent writes go to .sciencex.

Run the new layout long enough to verify providers, sessions, skills, agents, and scheduled tasks before manually archiving the old directory. ScienceX never deletes it automatically.