What problem does it solve?
Multi-agent workflows lack a structured way for agents to report state changes like blockers, completion, or disagreements, leaving operators without machine-readable visibility into coordination quality.
Core Features & Use Cases
- Structured Signal Emission: Write JSON signal files (hands-up, blocked, done, checkpoint, partnership) to a desk's .signals/ directory with ISO 8601 timestamps for dashboard ordering.
- Self-Assessment Scoring: Include intent, confidence, accuracy, and completeness scores so coordination quality is measurable rather than anecdotal.
- Outcome Calibration: Pair execution signals with reviewer outcome signals via a shared run_id to compute the honesty gap between self-assessed confidence and delivered quality.
- Use Case: An agent desk finishes a code review task and emits a done signal with self-assessment scores; a reviewer later emits an outcome signal with the same run_id, and the signals-dashboard displays the calibration gap.
Quick Start
Emit a checkpoint signal for the current desk by writing a JSON file to its .signals/ directory and noting it in the journal.