`Claude Code` adds dynamic workflows for parallel subagents

// related

• #0001

  #0001Agents & toolsPaperclip Releases1 hour ago

  `Paperclip` `v2026.529.0` adds document annotations and company skills CLI

  70radar

  PaperclipCoding agent platform — document and skill workflows

  Agent work now gets passage-level review threads and managed skill catalogs. Useful if you run self-hosted agent workflows; otherwise it is a strong product-direction signal.

  • —Issue documents now support revision-aware inline annotation threads with stable anchors, reducing vague review comments around agent-written specs.
  • —New skills CLI covers install, reset, audit, export, and assignment; skill provenance is tracked across bundled, catalog, runtime, and adapter sources.
  • —Fresh private self-hosted installs get a one-time first-admin browser claim, removing the invite bootstrap problem on LAN deployments.
  • —Claude Local can refresh its Anthropic model catalog from the UI, so new Claude models can appear without waiting for a Paperclip release.

  Source: github.com/paperclipai/paperclip/releases/tag/v2026.529.Read original →
• #0002

  #0002Agents & toolsClaude Code Releases13 hours ago

  `Claude Code` `v2.1.158` adds Auto mode on Bedrock, Vertex, and Foundry

  60radar

  Claude CodeCoding agent CLI — automates code work with Claude

  Auto mode now reaches managed-cloud deployments for Opus 4.7 and Opus 4.8. Useful only if you run through Bedrock, Vertex, or Foundry.

  • —CLAUDE_CODE_ENABLE_AUTO_MODE=1 enables the feature, so rollout is opt-in rather than default behavior.
  • —Support covers Bedrock, Vertex, and Foundry for Opus 4.7 and Opus 4.8; direct API users get no new signal here.
  • —Practical value is higher for teams keeping LLM traffic inside cloud-vendor rails; local CLI-only workflows can skip this update.

  Source: github.com/anthropics/claude-code/releases/tag/v2.1.158Read original →
• #0003

  #0003Agents & toolsGeekNews16 hours ago

  `SQLite` says no to agent-written code PRs

  50radar

  Agent-generated patches are treated as input, not merge-ready code. For upstream work, legal provenance now matters as much as implementation speed.

  • —AGENTS.md targets people using coding agents inside the codebase, setting explicit PR and generated-code rules.
  • —Pull requests are rejected without prior agreement or public-domain legal paperwork. Fast patches still need clean ownership.
  • —Actual changes are reimplemented by maintainers, so agent output works better as a proposal than as final upstream code.

  Source: news.hada.io/topic?id=30002Read original →