Skip to main content

Override for One Project

One project needs explicit, reproducible harness intent. This path scaffolds a project config at <project>/.agents/drwn/config.json, selects one Worker root, and writes project-scoped downstream state into <project>/.claude, <project>/.codex, and <project>/.cursor.

Prerequisites​

  • drwn installed and the machine already set up — see Set Up Your Machine.
  • A project directory you can write to.

Scaffold the project overlay​

drwn init writes <project>/.agents/drwn/config.json. Pass --non-interactive (or --minimal) when you want a prompt-free minimal config; the default is guided when both stdin and stdout are TTYs.

cd /path/to/project
drwn init
drwn init --non-interactive

init warns if your .gitignore appears to exclude .agents. It never edits .gitignore for you.

Edit the overlay​

The config is plain JSON. Use the project mutation commands to keep the config and lock graph coherent, and use capability-specific commands for explicit project overlays.

drwn add @team/operator@^1.0.0
drwn use @team/operator
drwn add skill <skill-name>
drwn add mcp <server-name>
drwn extensions add parallel
drwn extensions add markitdown
drwn extensions add beads --include-skill

The shape is documented in the project config schema. Every config carries schema: "drwn.project-config", schemaVersion: 1, an ordered workers root list, and one activeWorker name or null. Optional explicit overlays include skills, mcpServers, extensions, targets, and trustedSources.

Machine capability isolation inside configured projects. When a project config is present, project-safe packaged policy combines with the selected Worker closure and explicit project overlays. The profile and explicit capability IDs in ~/.agents/drwn/machine.json do not become project intent. User-home output may remain ambient in the downstream client and is reported separately. A different teammate can reproduce the declared project state from the project files plus immutable Card content.

Confirm the project overlay is detected​

drwn status
drwn status --explain
drwn status --json

The JSON output identifies schema: "drwn.project-status", the installed roots, one activeWorker, its active Card closure, explicit overlays, declared capabilities, and diagnostic-only ambient observations. If project state is missing, drwn did not find the config; re-check the path or run from inside the project directory.

For provenance questions ("why is this skill in the effective state?"), use drwn status --why:

drwn status --why skill:<name>
drwn status --why server:<name>

Preview, then write​

drwn write --dry-run
drwn write

drwn write materializes the selected root's aggregate plus explicit project overlays into <project>/.claude, <project>/.codex, and <project>/.cursor. It does not mutate project intent and does not project machine capabilities into the project. The project write record at <project>/.agents/drwn/write-record.json tracks what was written so the next run can clean up safely.

Verify​

drwn status
drwn doctor
ls <project>/.claude/skills
ls <project>/.codex/skills
cat <project>/.cursor/mcp.json

drwn doctor against the project (run from inside the project directory) reports project-scoped issues — unknown skill references, stale target overrides, card-skill availability, and the rest of the categories described in Reading Doctor.

Cross-References​