-
Notifications
You must be signed in to change notification settings - Fork 0
Implement single-task execution lifecycle (assign, execute, complete) #21
Copy link
Copy link
Closed
Labels
prio:criticalBlocks other work, must do firstBlocks other work, must do firstprio:highImportant, should be prioritizedImportant, should be prioritizedscope:medium1-3 days of work1-3 days of workspec:agent-systemDESIGN_SPEC Section 3 - Agent SystemDESIGN_SPEC Section 3 - Agent Systemspec:providersDESIGN_SPEC Section 9 - Model Provider LayerDESIGN_SPEC Section 9 - Model Provider Layerspec:task-workflowDESIGN_SPEC Section 6 - Task & Workflow EngineDESIGN_SPEC Section 6 - Task & Workflow Enginespec:toolsDESIGN_SPEC Section 11 - Tool & Capability SystemDESIGN_SPEC Section 11 - Tool & Capability Systemtype:featureNew feature implementationNew feature implementationtype:testTest coverage, test infrastructureTest coverage, test infrastructure
Milestone
Description
Context
Implement the full single-agent task lifecycle per spec 6.1. This covers the complete flow from task creation through assignment, execution, and completion, including all status transitions, artifact collection, and proxy overhead metrics.
Acceptance Criteria
Task Lifecycle
- Task status transitions enforced: CREATED → ASSIGNED → IN_PROGRESS → IN_REVIEW → COMPLETED (or BLOCKED/CANCELLED)
- Task assignment: agent receives task with full context and acceptance criteria
- Agent executes task using LLM + tools from the agent engine
- Status transitions validated (no invalid jumps, e.g., CREATED → COMPLETED)
- Artifacts collected during execution (files created, outputs generated)
- Task completion includes summary of work done
- Error handling: task failure with reason, timeout with configurable duration
Proxy Overhead Metrics (§10.5 M3)
- Cost tracked for the entire task lifecycle
-
turns_per_task: number of LLM turns to complete a task -
tokens_per_task: total tokens (input + output) consumed -
cost_per_task: total USD cost (derived from existingCostRecorddata) - Metrics available on task completion for analysis
Testing
- Unit tests for status state machine
- Integration test for full lifecycle (assign → execute → complete)
Dependencies
- Depends on Implement agent engine core with ExecutionLoop protocol integration (DESIGN_SPEC §3.1, §6.1, §6.5) #11 (agent engine core)
- Depends on Design and implement basic tool system (registry, invocation, results) #15 (basic tool system)
Design Spec Reference
- §6.1, §6.2 — Task Lifecycle and Workflow
- §10.5 — LLM Call Analytics (M3: proxy overhead metrics)
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
prio:criticalBlocks other work, must do firstBlocks other work, must do firstprio:highImportant, should be prioritizedImportant, should be prioritizedscope:medium1-3 days of work1-3 days of workspec:agent-systemDESIGN_SPEC Section 3 - Agent SystemDESIGN_SPEC Section 3 - Agent Systemspec:providersDESIGN_SPEC Section 9 - Model Provider LayerDESIGN_SPEC Section 9 - Model Provider Layerspec:task-workflowDESIGN_SPEC Section 6 - Task & Workflow EngineDESIGN_SPEC Section 6 - Task & Workflow Enginespec:toolsDESIGN_SPEC Section 11 - Tool & Capability SystemDESIGN_SPEC Section 11 - Tool & Capability Systemtype:featureNew feature implementationNew feature implementationtype:testTest coverage, test infrastructureTest coverage, test infrastructure