No. 013 · 20 August 2026 · 3 min← No. 012 · No. 014 →

SIGNALNORMAL|SHIPPED 2|SKIPPED 5DeepSeek peak/off-peak API rates liveRETIRING MAI-Code-1-Flash (Copilot) · 21d

The one thing

Anthropic Files API and Agent Skills are GA — drop the beta headers.

Platform release notes dated 19 Aug: /v1/files and Messages requests that reference an uploaded file no longer need files-api-2025-04-14. GA response shape adds expires_in_seconds / expires_at, page / next_page list pagination, and an ids[] filter. Org storage is 1 TB; rate limit 500 req/min. Beta-header callers keep the old response format.¹

Agent Skills and /v1/skills are GA the same day: no skills-2025-10-02 header, including Messages that load Skills via container. Anthropic Admin API user-management for Claude Enterprise (members, invites, groups, custom roles) is also GA — the ce-user-management-2026-07-13 beta header is optional. Python SDK 0.124.0 ships the Files/Skills GA surface plus computer-use and browser-use toolsets.¹²

VERDICT · USE IT — Pin anthropic ≥0.124.0, drop both beta headers on new code, and set file expiry on upload. Keep the beta header only if a client still parses the pre-GA list shape.

Deeper — Managed Agents extras same day

Managed Agents: allowed_domains / blocked_domains on web_search and web_fetch in agent_toolset_20260401 configs; web_fetch also takes max_content_tokens.¹
Self-hosted sandbox sessions can attach memory stores; SDK workers mount at mount_path and sync changes back.¹
Python SDK 0.125.0 adds Managed Agents web-search config and self-hosted sandbox memory types.³

Shipped

Use it

Claude Code 2.1.236–237 · default model env, idle notify, Concise style, gateway cache

ANTHROPIC_DEFAULT_MODEL sets the model new sessions start on; a /model pick still overrides and persists (unlike ANTHROPIC_MODEL). Cross-session SendMessage gains opt-in notify_when_idle (one-shot, macOS/Linux). macOS sandbox: wildcard read-denies (e.g. **/.env) win inside allowed regions and cover directory contents. 2.1.237 fixes prompt caching through LLM gateways / custom base URLs and adds a built-in Concise output style under /config.⁴⁵

MATTERS TO · Claude Code daily drivers on multi-session hosts, gateway proxies, and anyone drowning in preamble

nofeed.dev/issues/2026-08-20/claude-code-2-1-236-237/

Use it

openai-agents-python 0.22.0 · guardrail redaction + explicit-client contract

Minor version for runtime hardening. Terminal function-tool output blocked by output guardrails is redacted from replayable and persisted SDK state. Non-streaming Responses with terminal failed or incomplete raise ModelBehaviorError. If you pass an explicit openai_client with organization or project, those values must live on the AsyncOpenAI client — OpenAIProvider rejects the conflict. Usage accounting isolates across independent RunState checkpoints.⁶

MATTERS TO · OpenAI Agents SDK users with guardrails, multi-checkpoint runs, or explicit client wiring

nofeed.dev/issues/2026-08-20/agents-python-0-22-0/

Promised, not shipped
OpenAI Private Safety Processing / frontier ZDR detail — white paper expected September (vendor + secondary) · Origin agent-native source-control features — "ship soon" (issue 011) · Claude Code /design skill — research preview; no GA version line · MAI-Code-1-Flash removed from Copilot — 10 Sept 2026 · Codex send_user_message_async — merged to public repo; not announced as a general capability · Grok Bot enterprise — waitlist, no date · OpenAI Ultrafast (GPT-5.6 Sol, Cerebras) — limited preview expands with capacity

The conversation

The claim — including ours

Anthropic took Files and Skills out of beta the same window OpenAI reaffirmed frontier ZDR and previewed multi-session safety without retaining prompts.¹⁷

From the floor

Anthropic platform notes · 19 AugVendor, dated
Files API and Agent Skills GA; Admin user-management GA for Claude Enterprise; Managed Agents domain allow/block and self-hosted memory mounts.¹
OpenAI · Zero Data Retention (via TechCrunch + vendor index)Vendor claim; primary page timed out on crawl
Eligible API customers keep ZDR on frontier models; Private Safety Processing previews long-horizon misuse checks without retaining prompts — September rollout, eligibility unspecified on secondary coverage.⁷⁸
HN · Opus 5.0 comment noiseReader thread
Field complaint that Opus 5 over-comments even against CLAUDE.md; product lead reply generated with Claude Code — colour, not a bench.⁹

Our take

Ship Files/Skills without beta headers today. Upgrade Claude Code for gateway cache and Concise. Pin agents-python if you wire an explicit client or store run state. Treat OpenAI frontier ZDR as a September checklist item until the white paper names who qualifies.

GitHub releases: 28 watchlist repos, sweep_complete true, authenticated, 23 in-window (llama.cpp nightlies and AI SDK monorepo patches collapsed). Vendor feeds: 12 sources, sweep_complete true, 10 in-window. X accounts digest (roster ~100) and X keyword digest last collected 05:30 UTC 19 Aug (~24.5h old at window close) — background only. Reddit: fresh pulse (6 subreddits, 8 posts) plus 19 Aug 05:30 digest. HN: Opus 5 / fx threads. arXiv not swept. OpenAI ZDR primary page returned crawl timeout; facts held to TechCrunch quote of vendor + secondary confirmations.

Skip this

OpenAI ZDR / Private Safety Processing announcement. September white paper and eligibility still unpublished; secondary coverage only on the primary URL timeout.

Cursor Fable 5 outage 2.7h major; Composer 2.5 errors 3.0h minor; ChatGPT login 0.9h minor; Opus 5 / Haiku 4.5 1.3h minor. Resolved weather with duration — not a ship decision.

CodeQL 2.26.3 and org Code Quality Trends. Security/enterprise dashboard. Outside install-or-pin for most of this desk.

Replit Free Mode on GPT-5.6 Luna; LFM2.5 QAD checkpoints. Product/partner and model-weight notes, not a default-stack change.

llama.cpp b10505–b10507; AI SDK monorepo patches; openai-python 3.3.1 dep bump. Nightlies and gateway/security bumps.

Everything we saw

52 candidates scanned · 11 used in this issue — the rest, with the reason each one was left out

The receipt. Ranking is only trustworthy if the discarded pile is visible, so here it is: everything the collectors surfaced in the window, with its signal and what we did with it.

Item	Source	Signal	Call
Anthropic platform release notes — 19 Aug (Files/Skills GA)	anthropic docs	GA headers dropped	led the issue
anthropic-sdk-python v0.124.0	github releases	13.2h	one thing support
anthropic-sdk-python v0.125.0	github releases	8.1h	deeper
Claude Code v2.1.236	github releases	10.0h	shipped
Claude Code v2.1.237	github releases	5.2h	shipped
openai-agents-python v0.22.0	github releases	16.3h	shipped
OpenAI ZDR for frontier models (index)	openai.com	primary timed out	conversation / promised
TechCrunch — OpenAI Private Safety Processing	techcrunch	secondary	conversation
HN — Opus 5.0 incoherence / comments	news.ycombinator.com	thread	conversation colour
Cursor status — Fable 5 unavailable 2.7h major	cursor status	resolved	skipped weather
OpenAI status — ChatGPT login 0.9h minor	openai status	resolved	skipped weather
CodeQL 2.26.3	github changelog	security	skipped
X digests 19 Aug 05:30	digests	~24.5h old	background only
Reddit pulse — 6 subs, 8 posts	reddit pulse	layer ran	layer ran
Codex async messaging (The New Stack)	thenewstack	merged, not GA	promised

EOF

End of feed. That is everything from the window worth your time.
Explore more news, analysis and practical guides on NO FEED.

Still developingOrigin stays early beta (issue 011). DeepSeek peak/off-peak live (issue 010). Ultrafast expands with capacity (issue 009). Muse Glimmer runtimes still rolling (issue 006). Sonnet 5 docs lag $2/$10 (issue 006). MAI-Code-1-Flash → 1.1-Flash closes 10 Sept. DeepSeek Harness stays preview. Claude Cowork web/mobile remains beta. OpenAI Private Safety Processing / frontier ZDR detail due September.

Sources — each with the date it was read01 Anthropic — Claude Platform release notes (19 Aug 2026) (2026-08-20) · 02 github — anthropic-sdk-python v0.124.0 (2026-08-20) · 03 github — anthropic-sdk-python v0.125.0 (2026-08-20) · 04 github — claude-code v2.1.236 (2026-08-20) · 05 github — claude-code v2.1.237 (2026-08-20) · 06 github — openai-agents-python v0.22.0 (2026-08-20) · 07 OpenAI — Offering Zero Data Retention for frontier models (primary; crawl timed out) (2026-08-20) · 08 TechCrunch — OpenAI Private Safety Processing (secondary) (2026-08-20) · 09 HN — Opus 5.0 drives incoherence into the stratosphere (2026-08-20)