Claude Code subagents can nest since v2.1.172; agent-skills docs still say impossible

4 answers

• Best answerAgentArmor teamTeam4d ago

  Status: open as of this thread (July 2026). The repo owner agreed this is documentation drift, not wrong original guidance, and said PR #380 is up with a first cut at separating the repo's permanent convention from the Claude Code platform behavior that moved underneath it; the thread does not state whether that PR has landed.

  What actually changed in Claude Code: as of v2.1.172, subagents can spawn their own subagents up to 5 levels deep, so the repo's wording in AGENTS.md, docs/agents.md, and references/orchestration-patterns.md that nested subagents are impossible no longer matches current behavior. To still prevent nesting, a project can omit the Agent tool from an explicit tools allowlist or use disallowedTools. As of v2.1.178, Agent Teams no longer use TeamCreate/TeamDelete; with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, each session has one implicit team and teammates are spawned directly with the Agent tool's name parameter. Agent Teams still cannot nest: teammates cannot spawn their own teammates.

  The repo rule itself is unchanged: personas do not invoke other personas, composition stays with slash commands or the user, and Agent Teams remain the only endorsed pattern when teammates need to communicate directly. The issue also flags the Agent Teams worked example's separate 'Clean up the team' step, which reads like the old TeamDelete-style behavior; current docs describe graceful teammate shutdown and possible orphaned tmux cleanup instead. One collaborator noted the drift also trips the repo's source-driven-development skill, since an agent verifying claims against the official docs would find contradictions with current behavior.

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam4d ago

  What a GitHub user tried on GitHub (Partly worked)

  The reporter compared AGENTS.md, docs/agents.md, and references/orchestration-patterns.md against the official Claude Code sub-agents docs, Agent Teams docs, and changelog, concluding the notes predate the June 2026 updates (subagents can now nest up to 5 levels; Agent Teams no longer use TeamCreate/TeamDelete). The owner confirmed it was real drift, but no doc fix is confirmed in the thread.

  Source: https://github.com/addyosmani/agent-skills/issues/377 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam4d ago

  What @addyosmani tried on GitHub (Outcome unclear)

  The repo owner said PR #380 is already up as a first cut at separating the permanent convention (personas do not invoke personas) from the changed platform behavior, and asked the reporter to review it. Whether the PR fixes the docs is not stated in the thread.

  Source: https://github.com/addyosmani/agent-skills/issues/377#issuecomment-4937460547 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport
• AgentArmor teamTeam4d ago

  What @nucliweb tried on GitHub (Outcome unclear)

  A collaborator noted the drift also trips the repo's source-driven-development skill, because an agent verifying claims against the official docs would find contradictions with current Claude Code behavior. No fix was proposed in that comment.

  Source: https://github.com/addyosmani/agent-skills/issues/377#issuecomment-4929746139 (opens external site) ↗

  Did this work for you?WorkedDidn't work
  UpvoteStarReport