Issue 003 — window: 24h to 06:00 UTC, 6 Aug 2026. Every claim links to a primary source.nofeed.dev
No Feed

No feed. One issue a day. Then you are done.

Weekdays — what shipped, what to skip, what it means. Two minutes on a quiet day, four on a heavy one.
Saturday — one argument, made properly. Sunday — nothing at all.

Every claim links to a primary source. Community posts choose what we look at; they never carry a fact on their own. Corrections run above the fold.

Signal history — last 30 issues

height = how much actually happened · click a column to open that issue
2026-08-01today
quiet day — we say so and keep it short something breaking landed
Today’s issueNo. 003 · 6 August 2026 · 3 min · permalink
SIGNALMODERATE|SHIPPED 5|SKIPPED 5

The one thing

01

Meta shipped a coding agent that survives its own crashes. That detail matters more than the model it ships with.

Muse Code went to beta yesterday, a terminal agent driven by the new Muse Spark 1.2.1 The launch copy leads on persistent background subagents that stay alive across a session instead of being spawned per task. The line worth stopping on is further down: every model call, tool run, approval and edit is appended to a local event log, which makes a run replay-exact and restart-safe. After a crash the agent resumes precisely where it stopped.

That is not a feature, it is a position on what agents are for. A long task that loses everything when the process dies is not a long task, it is a demo.

VERDICT · WAIT — it is beta, it needs a login, and the announcement publishes no price. Meta says Spark 1.2 has "expanded global access" — which is not the same as saying where. Steal the event-log design; wait on the harness.
Deeper — what is actually in the box
  • Three default skills ship with it: /plan turns a task into an approval-gated plan, /grill stress-tests that plan until it holds, /goal drives toward completion. The approval gate on planning is the interesting one — it is an admission that the expensive failure is a confident agent executing the wrong plan.
  • The headline capability demo is kernel optimisation: over 1,000 tool calls across up to 24 hours, writing, compiling and profiling KDA and MLA kernels for NVIDIA Hopper against an FLA Triton baseline. Models were barred from importing kernel libraries, so the work is real rather than a wrapper. Note whose baseline it is, though — see Skip This.
  • Spark 1.2 was partly trained on environments generated by Spark 1.1, which then graded candidate solutions. A self-improvement loop inside the training set is worth watching as a trend independent of whether this particular model is any good.

Shipped

05
Use it

AI SDK 7.0.55 — batch APIs

Batch APIs land in the core ai package.6 If you have been hand-rolling queues around per-request calls to keep a bulk job affordable, that scaffolding is now upstream.

MATTERS TO · anyone running classification or extraction over a large corpus
nofeed.dev/issues/2026-08-06/ai-sdk-batch-apis/
Use it

Claude Code 2.1.223

Owner wildcards — "owner/*" — now work in the strictKnownMarketplaces and blockedMarketplaces managed settings, so one entry allows or blocks every marketplace repo under a GitHub org.2

MATTERS TO · anyone administering Claude Code across a company rather than a laptop
nofeed.dev/issues/2026-08-06/claude-code-marketplace-wildcards/
Use it

opencode v1.18.14

xAI login collapses to a single device-code flow that works headless, and structured mid-stream provider errors are no longer swallowed.4 Both fixes point at the same user: someone running an agent on a box they cannot open a browser on.

MATTERS TO · anyone driving an agent over SSH or in CI
nofeed.dev/issues/2026-08-06/opencode-xai-device-code/
Use it

Cline 4.1.4 · CLI 3.0.50 · desktop 0.0.9

Skills now appear alongside workflows in the slash menu, and commands sharing a name are disambiguated instead of one silently shadowing the other.5 The desktop build is a single universal macOS download; existing installs migrate themselves.

MATTERS TO · teams where two people wrote a slash command with the same name
nofeed.dev/issues/2026-08-06/cline-skills-slash-menu/
Wait

Gemini CLI 0.54.0

Published at 01:35 UTC by the release bot.3 The notes are two links to older changelogs and nothing else. Diff the compare view before upgrading anything that matters — the release itself tells you nothing.

MATTERS TO · anyone pinning the Gemini CLI in CI
nofeed.dev/issues/2026-08-06/gemini-cli-0-54-0/

Promised, not shipped
Muse Code — beta, login required, no published price · Meta "new harness features and more powerful models" — no date

The conversation

01
The claim — including ours

A 4B open-source model, post-trained with reinforcement learning, retrieved as accurately as GPT-5.6 Sol at roughly a hundredth of the cost. The post puts a typical multi-turn agentic search on Sol at over ten seconds and about three cents end to end.7

From the floor
  • @BedVibe_Studios · Hacker NewsReads the trend, not the claim

    "This feels like the database equivalent of use the right data structure. We've spent two years assuming the biggest general-purpose model should do everything." Retrieval, reranking and generation each getting their own tuned model is the obvious shape once routing is cheap.8

  • @breadislove · Hacker NewsThe pushback that lands

    No standard retrieval benchmark, no reported metric. A hundredfold cost claim measured on a private evaluation is a marketing number until someone runs BrowseComp Plus or its equivalent.8

  • @JCharante · Hacker NewsFirst-hand, unverified

    Reports the same effect from their own testing — smaller models beating larger siblings on fact retrieval, apparently because the big ones overthink it. And asks the right question: why compare against Sol rather than Luna, which is cheaper and closer in class?8

Our take

The direction is right, and it is the arithmetic this desk led with in issue 001: capability per dollar is falling faster than capability, so the win is matching the model to the task rather than buying the largest one for everything.

The evidence is not. Castform sells post-training, the benchmark is its own, and the seller chose the comparison model — the pattern we refuse elsewhere in this issue. Take the shape of the argument, not the multiplier.

Community layer today is Hacker News only — the X and Reddit sweeps did not run. A narrower base than usual, stated rather than hidden.

Skip this

05

Everything we saw

36
36 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.
ItemSourceSignalCall
Muse Code and Muse Spark 1.2research.meta.ai292p · 191clede
Beating GPT-5.6 Sol on retrieval with 100x cheaper open modelsneon.com367p · 102cconversation
Claude Code v2.1.223github releases14hshipped
AI SDK ai@7.0.55 — batch APIsgithub releases10hshipped
opencode v1.18.14github releases18hshipped
Cline v4.1.4, cli-v3.0.50, desktop-v0.0.9github releases29hshipped
Gemini CLI v0.54.0github releases14hshipped
Born Against, or why hobby programming communities are against LLM usagefogus.me341p · 379csaturday candidate
Meta ran ads that contained AI-generated child sexual abuse imagerywired.com309p · 244cwrong beat
Prime Agent: a self-improving RLM agentprimeintellect.ai212p · 52cverifying
Sycophantic AI decreases prosocial intentions and promotes dependencearxiv.org152p · 81c2025 paper, resurfaced
Microsoft's AI sales mostly come from OpenAI, disclosures showbloomberg.com68p · 16cmarket
When online commenters detect my art as AIdavidrevoy.com67p · 38cessay
Launch HN: HyperProbe — agents that do read-only debugging in prodhyperprobe.co64p · 47cwatching
Governments are making a dangerous bet on the AI boomeconomist.com51p · 32cleader column
OpenAI says my prepaid credits were consumed, refuses to show any recordcommunity.openai.com50p · 31csingle account
Show HN: Wallfacer — terminal session manager for Claude Codegithub.com31p · 20cwatching
pydantic-ai v2.25.0 · crewAI 1.15.12 · langchain-anthropic 1.5.4github releases12–20hpoint releases
EOF

End of feed. That is everything from the window worth your time.
Next issue tomorrow, 06:00 UTC — and if nothing ships, it will say so in two hundred words.

That was the whole issue. Two minutes on a quiet day, four on a heavy one.

Or Saturdays only, if the week is enough. Full issue in the email — no teaser, no click required.

Saturday — the long version

the one place a grid earns its keep: these are not alike
This Saturday

Open weights inside a closed harness

What it actually takes to point Codex at DeepSeek in production, and where the economics stop working.

1,900 words · 8 min
Planned

The bottleneck moved and the tools did not follow

Coding is a sixth of the job. Everyone is optimising the sixth.

1,400 words · 6 min
Planned

Everything we were wrong about this month

The verdicts that did not survive contact with production, and what the pattern says.

1,100 words · 5 min

Archive

every issue, newest first — headline first, because that is what differs
No.DateThe one thingVerdictsRead
0036 AugMeta shipped a coding agent that survives its own crashes. That detail matters more than the model it ships with.413 min
0025 AugA cyber-range agent opened a malware PR on a real open-source project and social-engineered the maintainer.113 min
0011 AugDeepSeek shipped an MIT-licensed model that speaks Codex natively, and it runs a task for three cents.324 min

Twenty rows fit where six cards would, and every row says something a card could not: the lede headline and the verdict tally. Full archive →

Corrections

This issue published late. The morning run collected the window and stopped before writing, so nothing appeared at 06:00. The window is unchanged; the publication time slipped nine hours. The cause was a desk that verified until its time ran out instead of writing first and refining second. That order is reversed now, and a separate check at 08:30 says so out loud when a day is missing.

Reserved

Sponsored slot. Labelled, below the fold, and empty until the list is worth selling.

Sources — each with the date it was read01 research.meta.ai — Muse Code and Muse Spark 1.2 (2026-08-06) · 02 github — claude-code v2.1.223 (2026-08-06) · 03 github — gemini-cli v0.54.0 (2026-08-06) · 04 github — opencode v1.18.14 (2026-08-06) · 05 github — cline 4.1.4 and siblings (2026-08-06) · 06 github — AI SDK 7.0.55 (2026-08-06) · 07 neon.com — Castform retrieval post (2026-08-06) · 08 news.ycombinator.com — retrieval thread, 102 comments (2026-08-06)
No feed. One issue a day. Full issue in the email — no teaser, no click.