← Reports
← Back to The Den
research

Agent Orchestration Research — Patterns & Repos

Generated 2026-04-02 08:03  ·  PKA / Lucienne

Research Sources

3 parallel researchers: Reddit/web, GitHub repos, architecture patterns. 2026-04-02.

Top Repos to Study

RepoStarsWhat to steal
BloopAI/vibe-kanban23.6kMCP dual-role (consume + expose), worktree per task, SQLite state
simple10/agents-observe-Claude Code hooks → WebSocket → React dashboard pipeline
hoangsonww/Claude-Code-Agent-Monitor-Kanban auto-moves via hooks, tool usage visualization
MeisnerDan/mission-control-ai-context.md condensed state file, Eisenhower matrix
23blocks-OS/ai-maestro575Agent-to-agent messaging protocol (AMP), peer mesh
CronusL-1141/AI-company145Task Wall (shared read surface), Loop Engine
saltbo/agent-kanban-Cryptographic agent identity, multi-runtime support
BradGroux/veritas-kanban-YAML workflow pipelines, enforcement gates
madviking/headless-pm-Pull-based lock/claim queue pattern
jayminwest/overstory-Full agent OS: sling/stop/trace/costs CLI

Architecture Patterns (from mature platforms)

ProblemPatternHow to implement
No real-time visibilityEvent bus (structured trace events)agent_events table in vault.db + WebSocket stream to dashboard
Unstructured resultsOutput schema contractsEach ticket type defines a Pydantic/TypedDict output schema
No follow-up ticketsspawn_ticket toolGive agents a create_ticket API call
No priority queueNumeric priority + age-based bumpPriority column on tickets, pickup sorts by priority DESC
No agent metricsThree-layer measurementagent_metrics table: duration, tokens, cost, quality score
No pause/cancelCheckpoint-based interruptcancel_requested flag on ticket, agent polls between steps

Key Insight

Events are the language, checkpoints are the safety net, schemas are the contract, and the bus decouples everything.

Our polling/cron model is sound for low-frequency work. The gaps are: missing event emission (hooks), output schemas, and cancel flags.

Priority Implementation Order

  1. Claude Code hooks → live dashboard — biggest visibility gap, agents-observe has the exact pattern
  2. ai-context.md generation — condensed workspace state for subagents, massive context savings
  3. Task Wall — shared view of all in-flight work across Den + MC
  4. Output schemas — structured results from agents, not raw text
  5. Agent metrics — duration, cost, quality per role