Who contributed to Anthropic on August 4, 2026?

5 developers shipped this update, including ashwin-ant, github-actions[bot], GitHub Actions, ant-kurt, and github-actions.

What were the notable Anthropic updates?

v0.2.129, fix(transport): validate skill names in ClaudeAgentOptions.skills, and Derive trigger timestamps for issues and pull_request events.

@anthropics

Claude SDKs and developer tooling

github ↗

Pick a date

Topics: AI / ML Full archive →

The Wire · Showcase

CLAUDE CODE AGENT SDK TIGHTENS SKILL VALIDATION, FIXES EVENT FILTERING

By RepoJournal · Filed 06:58 UTC on August 4, 2026 · About Anthropic

5 people shipped this

github-actions @github-actions 5 cited

ashwin-ant @ashwin-ant 2 cited

github-actions[bot] @github-actions[bot] 1 cited

ant-kurt @ant-kurt 1 cited

GitHub Actions @actions-user

The Python agent SDK now rejects malformed skill names that could silently disable tools, while the code action fixes inconsistent timestamp handling across GitHub event types.

Claude Agent SDK v0.2.129 ships a breaking change that validates skill names before they reach the CLI [1]. Names containing parentheses, commas, control characters, wildcards, or leading slashes now raise ValueError at connect time, closing a gap where malformed names could build rules that never match [2]. This lands alongside Claude Code v2.1.221, which adds a Focus view toggle to hide tool activity behind per-turn summaries [4]. On the automation side, claude-code-action now derives trigger timestamps for issues and pull_request events, ensuring comment and review filtering behaves consistently regardless of what triggered the run [3]. The CLI and plugin ecosystem are synchronized across all three releases, with knowledge-work-plugins rolling out five automated SHA bumps that passed validation before opening for review [5], [6], [7], [8], [9].

Action items

→ Update claude-agent-sdk-python to v0.2.129 and review any `skills=[...]` parameters for parentheses or special characters anthropics/claude-agent-sdk-python [immediate]
→ Review and merge knowledge-work-plugins SHA bumps in PRs 763-767 after scan validation anthropics/knowledge-work-plugins [plan]
→ Roll Claude Code to v2.1.221 to enable Focus view and sandbox credential masking anthropics/claude-code [plan]

References

[1] v0.2.129 ↗ anthropics/claude-agent-sdk-python
[2] fix(transport): validate skill names in ClaudeAgentOptions.skills ↗ anthropics/claude-agent-sdk-python
[3] Derive trigger timestamps for issues and pull_request events ↗ anthropics/claude-code-action
[4] v2.1.221 ↗ anthropics/claude-code
[5] bump(datarobot-agent-skills): 519c35ec → db831e2e ↗ anthropics/knowledge-work-plugins
[6] bump(figma): ef474d18 → 30b6fe64 ↗ anthropics/knowledge-work-plugins
[7] bump(valtown): d14627f6 → 0afe7b73 ↗ anthropics/knowledge-work-plugins
[8] bump(modern-web-guidance): ccb800c9 → 684ab9d7 ↗ anthropics/knowledge-work-plugins
[9] bump(wix): d33b1f64 → dadcea40 ↗ anthropics/knowledge-work-plugins

Quick answers

What shipped in Anthropic on August 4, 2026?: The Python agent SDK now rejects malformed skill names that could silently disable tools, while the code action fixes inconsistent timestamp handling across GitHub event types. In total, 7 commits, 9 pull requests, and 3 releases landed.
Who contributed to Anthropic on August 4, 2026?: 5 developers shipped this update, including ashwin-ant, github-actions[bot], GitHub Actions, ant-kurt, and github-actions.
What were the notable Anthropic updates?: v0.2.129, fix(transport): validate skill names in ClaudeAgentOptions.skills, and Derive trigger timestamps for issues and pull_request events.

CLAUDE SDK ROLLS OVERLOAD DETECTION AND SESSION RESUME CONTROLS

TypeScript Agent SDK v0.3.223 ships structural 529 overload detection and a new resumeDropsTurn safety valve for session truncation, while the Python SDK and Code Action synced their bundled CLI versions overnight.

ai 10 shipped

@anthropics Aug 5

CLAUDE CODE ACTION LOCKS DOWN WORKFLOW PERMISSIONS AND IMAGE PAIRING

The code action now enforces collaborator checks on workflow_run events and pairs downloaded images correctly by asset identifier instead of array position.

ai 107 shipped

@anthropics Aug 3

PLUGIN ECOSYSTEM ROLLS OUT VALIDATED UPDATES ACROSS THREE REPOS

Thirteen automated dependency bumps swept through the plugin pipeline overnight, each validated before merge across official, community, and knowledge-work repositories.

ai 41 shipped

@anthropics Aug 2

PLUGIN ECOSYSTEM ROLLS FORWARD WITH 46 VALIDATED UPDATES ACROSS THREE REPOS

The plugin validation pipeline ran full cycle overnight, pushing dependency updates through knowledge-work, official, and community registries with every bump pre-validated before merge.

ai 46 shipped

Elsewhere on the wire

AI Agents Aug 6

LANGCHAIN HARDENS ANTHROPIC TOOL CALLING AND GIT CREDENTIALS AS OPENHANDS SHIPS 1.10.0

LangChain fixed a critical credential leak in evaluation runs while Anthropic tool schemas get proper fallback handling, as three major agent frameworks push releases across the ecosystem.

ai-agents 54 shipped

Local LLMs Aug 6

OLLAMA FIXES KV CACHE CORRUPTION, LLAMA.CPP HARDENS FILE SEARCH, VLLM KILLS MOE LEGACY CODE

Ollama patched a critical cache corruption bug that corrupts lazy snapshots during multi-request prefills, while llama.cpp hardened Windows junction traversal and vLLM removed 14 hardcoded MoE shards across quantization methods.