Orchestrate parallel AI agents across office projects (invoices, contracts, reports, email) from a single workspace — the way polyforge manages dev repos.
For office workers and ops teams running Claude Code
across multiple projects simultaneously.
Run ./scripts/sessions.sh to open parallel agent
sessions — one per project.
./scripts/status.sh # See all managed projects
./scripts/sessions.sh # Parallel CC sessions
./scripts/run.sh "task" ./dir/ # Run task across projectsProjects: edit config/projects.conf. Credentials:
copy config/env.example; its names match the ${VAR} references in
mcp/*.json.
Both scripts start Claude with --add-dir pointing at this repo, so the
skills load in every project. run.sh runs headless with
--permission-prompts none and prints JSON (.result, .is_error,
.total_cost_usd) on stdout:
| Env var | Default | Effect |
|---|---|---|
OFFICE_FORGE_PERMISSION_MODE |
acceptEdits |
run.sh permission mode |
OFFICE_FORGE_MAX_BUDGET_USD |
5 |
run.sh USD cap per task |
- Skills — deterministic office workflows:
doc-organize,email-triage,invoice-processing(afinancial-reportskill is planned, #6; contract support is atemplates/client-engagement/contracts/scaffold only, not a skill) - MCP servers — business API configs (accounting, CRM, email)
- Templates — project scaffolds for common scenarios
- Business-process baseline — how the office-loop (business-process / IT) maps to the Startup CTO Handbook (estate-level note in qte77)