Fetching from the wire…
Public story · 2026-07-27 · high
Pairing the new agent-memory-2026-07-22 header with managed-agents-2026-04-01 now returns a 400, and list calls silently stop honoring order_by.
Why now: The header requirement and the path_prefix change are both documented in Anthropic's platform docs as covered in the July 27 briefing.
Claude's memory-store API requires a new beta header, agent-memory-2026-07-22, per Anthropic's docs.
The loud failure here is easy to catch, since it throws an error before anything ships. The quiet one isn't: it changes what a working query returns, with nothing in the response to flag it.
Combining the new header with the older managed-agents-2026-04-01 header on one request triggers the loud failure, a 400 instead of a merge.
The quiet change is to path_prefix. Say a filter used the prefix 'user' expecting it to catch paths like 'users/123' too. That stops working: the prefix has to end with a slash and match a whole path segment, not a fragment of one.
Listing calls changed too. They return a stable, server-defined order and ignore any order_by or order parameter sent with the request.
The depth parameter shrank too. It only accepts 0, 1, or being left out entirely.
None of this throws an error, not for path_prefix, not for listing order. That's the actual risk: a 400 gets caught in staging, but a quietly different set of memories doesn't get caught anywhere at all. Anyone with a path_prefix filter built on old substring behavior should check it against real path segments before it changes what an agent remembers.
Each link below shares sources, entities, or timing with this story.
The majority threshold crossed and nobody threw a party. Felo's analysis of 2026 developer tool data reports that over 51% of all code committed to GitHub in early 2026 was either generated or substantially assisted by AI tools. That's not a projection. That's the current stat...
The Quantization-Aware Healing post claims a compressed 4-bit model outperforms the uncompressed one. Technical readers clarified the actual claim: a 120B cut to a 60B BF16 model, then to 60B mxfp4 that beats the 60B BF16 but not the 120B base. Different claim entirely. The sh...
(intermediate) — Add memory: user or memory: project to subagent frontmatter. Agent gets auto memory directory. First 200 lines of MEMORY.md injected into every session. Code reviewer that remembers your patterns, debugger that recalls past fixes. (Claude Code Docs)
Today's release means an agent no longer carries the full tool surface in context, and CLI tool access becomes permission-scoped. A Tape Trace Inspector records agent runs, provider requests, tool calls and Skill usage, Skills unify into a shared library with per-agent enablem...
arXiv 2608.23541 tested 11 verifier-scored optimization tasks under matched compute. Different model families do find structurally different solutions, and then a single round of reading each other's complete outputs erases exactly the diversity that justified using multiple m...
The "Think in Code" rule took 47 Read() calls totaling 700KB down to one script returning 3.6KB (GitHub). You can apply this today with nothing installed, just by changing how you phrase multi-file questions.
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.