Independent project / Conductor
Make the dependencies visible.
Tracks, dependencies and task-specific context make parallel agent work easier to inspect before execution.
The problem
Agent work needs an explicit plan: which tasks can start, which depend on another track, and what context belongs to each task.
What I built
I built Conductor to organize agent work into tracks with dependencies and task-specific Markdown context. Its planning interface makes those relationships visible before execution.
Decision & tradeoff
A track carries its own context and dependencies instead of relying on one undifferentiated conversation. The scheduler checks the dependency graph before starting work, including dependency-cycle detection.
The plan is only as useful as the dependencies and context supplied to it. Describing scope in context is not proof of permission enforcement.
How the workflow fits
- Define tracks. Give each task a goal, context and declared dependencies.
- Resolve dependencies. Validate the graph and detect cycles before scheduling.
- Schedule eligible work. Start work according to dependency readiness.
Explore the implementation
Real local planning interface · three synthetic tracks · no workers executed.
1. Prerequisites
Component system and Retrieval API are separate prerequisite tracks in this synthetic example.
2. Dependency
Review interface depends on both prerequisites. The graph makes that relationship inspectable.
3. Execution boundary
This capture shows planning only. No agent workers executed, so the graph is not evidence of successful delivery.
Result & limits
The implementation validates declared dependencies, detects cycles and schedules using dependsOn. The captured interface makes the relationship between two prerequisites and a review track readable.
This demonstrates planning, not agent delivery, reliability metrics or employer adoption.
Inspect Conductor on GitHub ↗More engineering stories