| name | autonomous-execution |
| description | Reduce manual intervention — autonomous natural language execution triggers, delegation observability, and run persistence. |
Autonomous Execution
Natural language should trigger ARAYA execution. Commands become optional.
Delegation progress is visible. Run records persist for audit.
IMP-01: Natural Language Triggers
ARAYA recognizes intent from natural language and starts the appropriate
workflow without /araya run:
| Natural Language | Triggers |
|---|
| "Manu, assess mahg-pms" | PO assessment workflow |
| "Sonia, review this project" | Delivery review workflow |
| "Aurora, identify capability gaps" | GAR generation |
| "I want to reconstitute mahg-pms" | Reconstitution workflow |
| "Help me prepare a release" | Release readiness workflow |
IMP-02: Delegation Observability
During execution, The Professor sees:
- Current run ID
- Current phase
- Current agent
- Completed tasks / total tasks
- Blocked tasks (if any)
IMP-03: Run Persistence
Every run creates .araya/runs/{run_id}/run.json:
- Run ID, start/end time, workflow, agents, status, artifacts, errors, outcome
IMP-04: Provider-Aware Routing
Before delegation, check: rate-limit history, recent usage, known failures.
Avoid selecting providers near TPM quota.
IMP-05: Usability Gate
A feature is complete only when it can be used, solves a real problem,
can be demonstrated end-to-end, and reduces manual work.