Skip to content

docs: Claude Code v2.1.317 - email-leak prompt + fail-closed SDK hooks tighten trust boundaries; computer use GA on Bedrock/Foundry; FlashProve repo awaits source decision - #1433

Merged
claude-yolo[bot] merged 1 commit into
mainfrom
docs/claude-code-v2.1.317-20261103-0721
Nov 3, 2026
Merged

claude-yolo[bot] merged 1 commit into
mainfrom
docs/claude-code-v2.1.317-20261103-0721

Conversation

@claude-yolo

@claude-yolo claude-yolo Bot commented Nov 3, 2026

Copy link
Copy Markdown
Contributor

Night shift report

WHY THIS MATTERS: v2.1.317 is a big release with about 90 changelog lines, and much of it is about trust boundaries and admin control. Claude Code now asks before a web or shell tool call sends your account email address to the network. /schedule now warns when a cloud task picks up connectors you didn't ask for. Agent SDK hook callbacks can now fail closed with onFailure: "block". Gateway admins get a real cost-reporting API. On the platform side, the computer use toolset went from beta to stable on Claude Platform on AWS, Amazon Bedrock and Microsoft Foundry. The discovery probe also found a new public Anthropic repo we don't archive: anthropics/FlashProve. Adding it as a source is a human decision.

HIGHLIGHTS:

  • New repo found: anthropics/FlashProve. It was added to discovery.json under github_org_anthropics.not_fetched, which means Anthropic published a new public repo. Someone needs to decide whether it belongs in sources.json. No domain or review changes. tombstones.json changed only its date: no new deaths and no resurrections.
  • Security and privacy changes in v2.1.317:
    • A check now asks before a web or shell tool call sends the account email to the network.
    • /schedule warns when a new cloud task gets connectors you didn't ask for.
    • Remote Control and SDK hosts no longer offer "don't ask again" on a mod's tool calls.
  • Agent SDK: hook callbacks accept onFailure: "block", so a hook that crashes blocks the action instead of letting it run. The TypeScript docs add a slimmer /core entry point and a sandbox_required_unavailable startup failure reason (for when failIfUnavailable is on, needs SDK v0.3.314+).
  • Claude apps gateway admin:
    • New GET /v1/organizations/analytics/cost gives past spend by developer, day or IdP group.
    • A daily record of each developer's IdP groups is kept for admin.identity_retention_days.
    • OTel gets anthropic.plan, anthropic.seat_type, anthropic.provider and anthropic.custom_base_url attributes.
  • Behavior changes to know about:
    • The Read tool's "file unchanged" stub is removed, so re-reads return full content.
    • Subagents and teammates on a different model now use that model's own effort level.
    • claude -p inside a cloud or self-hosted runner session no longer shares the parent session's ID. CLAUDE_CODE_DISABLE_NESTED_SESSION_ID=1 restores the old behavior.
    • claude -p now exits 1 when transcript writes fail because the disk is full.
  • Computer use is stable on the cloud providers. computer_toolset_20260801 moved from Beta to Available on Claude Platform on AWS, Bedrock and Foundry. The 5.5-model docs now say plainly that computer_20251124 returns a 400 on the Claude API, while Bedrock still accepts it. The whats-new and migration pages for Haiku, Sonnet and Opus 5.5 were updated to match.
  • LLM gateway protocol: new rules for the automatic retry when an upstream rejects an anthropic-beta value by name, including a table of how long a rejected value stays out (until exit, for the rest of the conversation, or for that model).
  • Connectors, Enterprise Managed Auth: a second assertion issuer, Claude-issued (beta, https://identity.anthropic.com/issuers/<id>), alongside IdP-issued. MCP server operators get new metadata requirements: issuer must match exactly and token_endpoint must be on the same registrable domain.
  • Compliance API: a new "Invalid file ID" error, plus guidance to choose between the claude_file_ and claude_gen_file_ endpoints by the ID's prefix.
  • Env vars:
    • New to the docs: CLAUDE_CODE_OAUTH_REFRESH_TOKEN and CLAUDE_CODE_OAUTH_SCOPES (headless sign-in provisioning), CLAUDE_CODE_GLOB_TIMEOUT_SECONDS, CLAUDE_CODE_DISABLE_CFC_PROMPT and FORCE_HYPERLINK.
    • Also changed: Desktop SSH ConnectTimeout guidance, the Desktop Worktree toggle, and per-user columns in Console analytics.
  • Also included: a cyber verification program support article (per-platform enrollment for AWS, GCP and Foundry), Managed Agents Java/C# examples rewritten for typed SDK accessors, a new latency_optimization cookbook, and cookbook notebook refreshes.

Created by night-shift claude-yolo
Day-shift claude-yolo will review and merge this in the morning

…lure block, computer use GA on Bedrock/Foundry, new repo anthropics/FlashProve

Co-Authored-By: claude-yolo[bot] <claude-yolo@lroole.com>
@claude-yolo claude-yolo Bot changed the title docs: Claude Code v2.1.317 - gateway cost report API, SDK hooks onFailure block, computer use GA on Bedrock/Foundry, new repo anthropics/FlashProve docs: Claude Code v2.1.317 - email-leak prompt + fail-closed SDK hooks tighten trust boundaries; computer use GA on Bedrock/Foundry; FlashProve repo awaits source decision Nov 3, 2026
@claude-yolo
claude-yolo Bot merged commit 8a75ab5 into main Nov 3, 2026
1 check passed
@claude-yolo
claude-yolo Bot deleted the docs/claude-code-v2.1.317-20261103-0721 branch November 3, 2026 07:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

0 participants