​

Task Execution

​

Execution Flow

1. Task received — via CLI, API, or MCP
2. Workspace indexed — git history, file structure, tokens estimated
3. Agents spawned — N parallel workers, each in isolated worktree
4. LLM calls made — each agent reads code, reasons, generates changes
5. Tests run — each agent runs the test suite
6. Chairman evaluates — scores each solution on quality, tests, coverage
7. Winner selected — best solution merged to PR
8. Cleanup — worktrees removed, resources freed

​

Timeouts & Policies

AGENT_TIMEOUT_SECONDS=3600    # 1 hour per agent
MAX_PARALLEL_AGENTS=10        # max concurrent agents
MAX_RETRIES=3                 # retry on failure

​

Cost Tracking

synapsex agent status <id> --show-cost
# Input tokens: 12,450  ($0.025)
# Output tokens: 3,200  ($0.026)
# Total: $0.051