A coordination skill family for multi-agent task work in shared repositories.
Agent Foreman lets multiple AI agents (Claude, Codex, Cursor, etc.) work on the same codebase without stepping on each other. It uses the file system as the coordination layer — no external services, no databases, just YAML files in your repo.
Core workflow: discover → scan → claim → start → complete
npx skills add victorGPT/agent-foreman-skillsOr install a single skill:
npx skills add victorGPT/agent-foreman-skills --skill coordination-scan| Skill | Purpose |
|---|---|
agent-foreman |
Entry point — routes to the right sub-skill |
coordination-discovery |
Finds coordination setup, planning source, execution layer |
coordination-bootstrap |
Initializes coordination structure for new projects |
coordination-scan |
Reads task landscape — what's available, what's active, touch conflicts |
task-claim |
Claims an open task (open → claimed) |
task-start |
Marks task as actively in progress (claimed → in_progress) |
task-complete |
Closes a task with evidence (in_progress → done) |
open → claimed → in_progress → done
Active tasks live in tasks/<domain>/active/. Completed tasks move to tasks/<domain>/done/. Conflict detection is a single Glob — no file reads needed.
Add to your AGENTS.md:
# Coordination pointers
- Coordination root: docs/coordination
- Global execution index: docs/coordination/index.yaml
- Domain indexes root: docs/coordination/indexes
- Task records root: docs/coordination/tasksRun coordination-bootstrap to initialize the structure if it doesn't exist yet.