Workspaces and Project Context
The legacy CrabCode docs spread workspace facts across sessions, config, data usage, and CLI pages; this page gathers those facts into one entry point.
Source facts
CrabCode uses the launch directory as the project context. Session files are grouped by a sanitized project path, /resume defaults to the current project, and transcripts are stored locally as JSONL files.
What a workspace carries
- The current directory plus any extra directories explicitly added with
--add-dir. - Session history, file-change traces, and resume/rewind entry points.
- Project-level agent definitions:
<repo>/.crabcode/agents/<name>.md. - Repository-owned conventions such as hooks, MCP config, or custom instruction files.
Read next
- Resume, fork, rewind, and export: Sessions.
- Config paths and settings: Config reference.
- Local data boundaries: Data usage.