141 wires and counting

$ follow Anthropic

Keep up with Anthropic in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.

or

fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.

Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.

$ status

wire 2026-08-18
stories 46

© 2026 RepoJournal Home Showcase How it works Privacy

$ the-wire · showcase

SDK Python SDK Gains Typed Errors, Parent Tool Recovery, and Subagent Text Forwarding

By RepoJournal · Filed · About Anthropic · Composed from the cited sources · methodology

The Python Agent SDK just fixed three long-standing bugs and added a new option, all in a single day's work from qing-ant.

The Claude Agent SDK for Python shipped a stack of fixes overnight, led by a change that raises a typed `ResultError` with the real error-result text and payload [1]. This repairs three issues: untyped exceptions on the message-stream path, lost subagent parent IDs [2], and the new `forward_subagent_text` option that mirrors the TypeScript SDK's behavior [3]. Also fixed: `can_use_tool` now stays available with string prompts [4]. Meanwhile, the security plugin reached v0.10.1-rc7 [5], and automated bumps rolled through both plugin repos, including `exa` [6] and `dash0` [7]. Notably, the knowledge-work plugins repo saw a swath of bumps for `modern-web-guidance` [8], `carta-investors` [9], `wix` [10], and `tinyfish` [11]. Release notes for the merge in zeta-23-lean promise "one comparator statement set, at the paper's constants, in a self-contained Challenge.lean" [12]. 39 PRs and 7 commits across 4 repos: this is a busy morning.

Action items

References

  1. [1] Raise a typed ResultError with the real error-result text and payload ↗ anthropics/claude-agent-sdk-python
  2. [2] Recover parent_tool_use_id when reading subagent transcripts ↗ anthropics/claude-agent-sdk-python
  3. [3] Add forward_subagent_text option ↗ anthropics/claude-agent-sdk-python
  4. [4] Keep stdin open for can_use_tool and allow it with string prompts ↗ anthropics/claude-agent-sdk-python
  5. [5] Claude Security Plugin - v0.10.1-rc7 ↗ anthropics/claude-plugins-official
  6. [6] bump(exa): e64c11f2 → c42a8a0b ↗ anthropics/claude-plugins-official
  7. [7] bump(dash0): c8fe573a → 106fddfc ↗ anthropics/claude-plugins-official
  8. [8] bump(modern-web-guidance): 9e70fa4c → 460e5536 ↗ anthropics/knowledge-work-plugins
  9. [9] bump(carta-investors): 570f794b → 37b9d30c ↗ anthropics/knowledge-work-plugins
  10. [10] bump(wix): 2c342a53 → a04e0e58 ↗ anthropics/knowledge-work-plugins
  11. [11] bump(tinyfish): 80ce9bfa → ca1f3f6d ↗ anthropics/knowledge-work-plugins
  12. [12] Merge the challenge files into one main-results statement set ↗ anthropics/zeta-23-lean

Quick answers

What shipped in Anthropic on August 18, 2026?
The Python Agent SDK just fixed three long-standing bugs and added a new option, all in a single day's work from qing-ant. In total, 7 commits and 39 pull requests landed.
Who contributed to Anthropic on August 18, 2026?
5 developers shipped this update, including rfurman, eeeasley, Michael Moore, github-actions, and qing-ant.
What were the notable Anthropic updates?
Raise a typed ResultError with the real error-result text and payload, Recover parent_tool_use_id when reading subagent transcripts, and Add forward_subagent_text option.