Claude Code Subagents & Orchestration (24)

Introduces Triglav, an open-source workflow where a Staff agent plans, Senior agents implement, a shared board tracks decisions and layered review checks Claude Code and Codex work.

Introduces ccdeck, a local dashboard (npx ccdeck) that draws each Claude Code session and its subagents as a live tree of cards showing state and recent tool calls.

The author measured a Claude Code usage week and found subagents, not the plan cut, were eating 48% of it.

A playbook for running 10-15 parallel Claude Code sessions by treating them like a small distributed system with clear ownership.

Cost comparison of three custom subagents in .claude/agents/ that each set a different model in their frontmatter.

What Claude Code's per-session and per-subagent JSONL transcripts reveal when you try to count your agents' activity.

Lessons from three months running a team of Claude Code agents on a production codebase, including agents grading their own work.

Four silent failure modes found when fanning a design-token sweep out to parallel Claude Code subagents.

A four-agent Claude Code pipeline that takes one command and works through implementation unattended.

How to keep context and decisions consistent across several Claude Code agents working in separate git worktrees.

How to use the expensive Fable 5 model only for judgment-heavy work while cheaper subagents handle implementation and chores.

How the author set up Claude Code with 26 production subagents, CLAUDE.md, MCP servers and hooks.

Giving Claude Code a team: an issue maintainer, a no-code orchestrator and specialist subagents that build in parallel.

A single command that sets up a reusable crew of Claude Code subagents for each new project.

Agent-pd, a zero-token audit log for catching subagents that go off course.

Rayline, a gateway that routes Claude Code subagent calls to on-device or cheaper models.

Lessons learned running large swarms of coding agents with a central orchestrator.

An experiment letting Claude Code and Codex talk to each other in real time through Git.

Anthropic's introduction of Dynamic Workflows for running large Claude Code tasks as planned steps.

Dari-docs, which uses parallel coding agents to test and optimize documentation for AI agents.

adamsreview, a Claude Code plugin for multi-stage PR reviews using parallel subagents and validation passes.

A drop-in CLAUDE.md with 11 rules (worktree isolation, file locks, depth caps) to stop parallel agents conflicting.

A solo developer's architecture for running 10 Claude Code and 2 Codex instances in parallel.

Using Claude Code to build a large organization of AI agent roles for team workflows.