Persistent project roadmaps for OpenCode. Coordinates work across sessions and parallel Task tool subagents.
OpenCode's built-in todo is session-scoped—it disappears when you restart. Task tool subagents are stateless—they can't see each other's work.
This plugin solves both:
- Persists to disk — survives restarts, available across sessions
- Shared context — subagents read the same roadmap to understand the bigger picture
- Concurrent awareness — agents see what's
in_progressand avoid conflicts
Add to your opencode.json:
{
"plugin": ["@howaboua/opencode-roadmap-plugin@latest"]
}OpenCode installs it automatically on next launch.
Create or extend a project roadmap. Requires a feature list and a short spec for each feature.
"Create a roadmap with features: 1) Auth, 2) Profiles. Specs: Auth uses OAuth and must support password reset; Profiles needs avatar uploads and privacy settings"
- Features group related work (
"1","2","3") and include a brief spec - Actions are markdown task list items (
- [ ] 1.01 ...) within features - New actions always start as
pending - Append-only: existing IDs never change
View current state and progress.
"Show me the roadmap"
"What's the status of feature 2?"
Before delegating work to Task tool subagents, instruct them to read the roadmap first so they understand their assigned action within the broader plan.
Change action status or description. Each update includes a brief note appended to the updates section.
"Mark action 1.01 as in_progress — Drafted schema notes"
"Action 2.03 is completed — Added tests for edge cases"
Statuses: pending → in_progress → completed | cancelled
Transitions are flexible—you can revert if plans change. Only cancelled is terminal.
Auto-archives the roadmap when all actions reach completed.
When multiple subagents work simultaneously:
- Each reads the roadmap to see what's
in_progress - Agents stay focused on their assigned action only
- They avoid modifying files that belong to another
in_progressaction - Errors outside their scope get noted, not fixed
This prevents conflicts when subagents run in parallel.
You: "Plan out building a REST API with auth, users, and posts endpoints"
AI: Creates roadmap with 3 features, ~12 actions
You: "Implement feature 1"
AI: Reads roadmap → sees Feature 1 has 4 actions → uses todowrite for immediate steps → delegates to subagents → each subagent reads roadmap first → updates status when done
- Active: Stored as a markdown roadmap alongside the project
- Archived: Snapshot archived when complete
MIT