July 30, 2026
Report summary
10 stories cleared the bar, led by Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident, Codex Security, and Discovering Cryptographic Weaknesses with Claude.
Worth attention
A detailed technical postmortem of a real July 2026 security incident involving an intrusion via a frontier AI lab's coding agent, including the attack timeline and how it was detected. Directly relevant to anyone running autonomous coding/agent workflows with broad tool access — worth reading to check your own agent permission boundaries and logging.
OpenAI has published a dedicated security repo/guidance for its Codex coding agent, covering sandboxing, permission boundaries, and safe-usage practices. Directly relevant if you run Codex (or similar coding agents) with filesystem/shell access — worth reviewing your own agent permission setup against it.
Anthropic published research demonstrating Claude being used to discover real cryptographic weaknesses/vulnerabilities. A concrete, vendor-verified example of using an AI agent for practical security research — directly useful if you want to apply the same approach to auditing your own projects (e.g. Veremun/ContractorVerify's compliance-sensitive code).
Practical write-up on using Nix to build and package lower-level systems software, covering pain points and patterns for reproducible builds. Useful reference for anyone considering Nix for deployment reproducibility, though it's a narrow tooling choice.
A well-regarded ML educator (Sebastian Raschka) breaks down the architecture of the new Kimi K3 open model — attention design, MoE structure, and training notes. Useful if you're evaluating open-weight models for local or self-hosted inference.
Open-source project (slurpjson) demonstrating GPU-accelerated JSON parsing using compute shaders, with working code on GitHub. A niche but concrete performance-engineering technique worth knowing about if JSON parsing throughput ever becomes a bottleneck.
An accessible technical explainer walking through the intuition behind Kimi's 'delta attention' mechanism, aimed at building understanding of a newer attention-architecture variant. Useful background if you're evaluating or fine-tuning models that use this approach, but educational rather than immediately actionable.
Argues that writers relying solely on Substack should also own a personal website/domain for control, SEO, and portability, with practical reasoning about platform risk. Relevant to anyone running a public-facing newsletter or content site, including your own librarian digest site.
An open-source notetaking app explicitly designed to be used by both humans and AI agents as a shared memory/context store. Conceptually adjacent to agent-memory/second-brain systems — worth a look for ideas or as a potential integration, though it's an early-stage project.
A video interview with Boris Cherny, creator of Claude Code, likely covering the tool's design philosophy and roadmap direction. Worth watching for anyone building on or heavily using Claude Code / agent-based dev tooling, though it's a video with no transcript summary available yet.
Full digest
Personal essay arguing Rocq (formerly Coq) remains preferable to Lean for formal program verification work, pushing back against the growing popularity of Lean. Niche academic tooling debate with no bearing on typical solo-builder workflows.
Statistics-focused post arguing the arithmetic mean is frequently a misleading summary statistic, with examples of when medians or other measures serve better. General audience math writing, not software-specific.
Technical deep-dive on multi-seat Wayland setups (multiple users/cursors on one Linux machine), covering current tooling gaps and workarounds. Interesting Linux desktop internals but a narrow hobbyist use case.
Blog post reflecting on how the choice of developer tools shapes what an engineer builds. General musing without concrete recommendations or data.
Personal essay celebrating the fun of building small, low-stakes 'toy' software projects outside of professional pressure. Reposted 2025 piece with no new information.
Practical write-up on using Nix to build and package lower-level systems software, covering pain points and patterns for reproducible builds. Useful reference for anyone considering Nix for deployment reproducibility, though it's a narrow tooling choice.
Essay questioning the fuzzy definition of 'microservices' and how the term gets used inconsistently across the industry. General architecture commentary without new data.
Conference talk from Richard Feldman on how different programming communities develop different norms around dependency usage. Thoughtful but a video talk with no transcript/summary provided in the feed.
Open-source project (slurpjson) demonstrating GPU-accelerated JSON parsing using compute shaders, with working code on GitHub. A niche but concrete performance-engineering technique worth knowing about if JSON parsing throughput ever becomes a bottleneck.
A detailed technical postmortem of a real July 2026 security incident involving an intrusion via a frontier AI lab's coding agent, including the attack timeline and how it was detected. Directly relevant to anyone running autonomous coding/agent workflows with broad tool access — worth reading to check your own agent permission boundaries and logging.
Essay on the lack of informal community spaces ('third places') for people learning to code or other skills outside formal institutions. General social commentary, not software-specific.
An accessible technical explainer walking through the intuition behind Kimi's 'delta attention' mechanism, aimed at building understanding of a newer attention-architecture variant. Useful background if you're evaluating or fine-tuning models that use this approach, but educational rather than immediately actionable.
A hobby project building a Linux desktop environment entirely from scratch. Interesting systems-programming showcase but an early-stage hobby OS project with no user-facing maturity yet.
Meta post about the Lobsters community site's hosting/infrastructure situation, referencing a piece on 'what happens after you IPO.' Site-operations trivia with no external relevance.
Deep technical reverse-engineering writeup on how IBM i (AS/400) stores and hashes passwords via the undocumented QSYRUPWD mechanism. Rigorous security research, but scoped to a legacy enterprise platform far outside a typical solo builder's stack.
A hobby emulator project for the retro CP/M-68K operating system. Fun retrocomputing work, purely a nostalgia/hobby item.
A technical post exploring how to encode 'proof types' (computational witnesses) using final classes in Dart's type system. Niche language-design technique for a language most solo builders in infra/backend work don't use.
A fun hack showing how to run Tailscale on a jailbroken Kindle to proxy traffic. Entertaining hobbyist networking trick, not a work-relevant technique.
A retrospective look at UI design in the demoscene (real-time graphics/audio art community). Interesting design history but pure entertainment/nostalgia content.
OpenAI has published a dedicated security repo/guidance for its Codex coding agent, covering sandboxing, permission boundaries, and safe-usage practices. Directly relevant if you run Codex (or similar coding agents) with filesystem/shell access — worth reviewing your own agent permission setup against it.
A personal userscript that merges an HN-linked article and its comments discussion into one resizable side-panel view. Handy personal productivity tool but very narrow scope.
Launch announcement for LearnVector, a new AI education company from Andrew Ng focused on one-to-one personalized learning. Notable founder, but a consumer ed-tech product with no direct relevance to a solo software builder's tooling or business decisions.
Argues that writers relying solely on Substack should also own a personal website/domain for control, SEO, and portability, with practical reasoning about platform risk. Relevant to anyone running a public-facing newsletter or content site, including your own librarian digest site.
A retro-computing project porting the 1998 game Half-Life to run natively on classic Mac OS 9. Impressive engineering feat but pure nostalgia/entertainment content.
Product launch for an e-paper display camera/photo device. Consumer hardware gadget, not software/dev-relevant.
A theoretical critique arguing that 'linear probe' techniques for detecting truthfulness inside LLM activations are conceptually flawed, drawing on Tarski's semantic theory of truth. Interesting interpretability research but abstract and not tied to a reproducible tool or actionable change.
A well-regarded ML educator (Sebastian Raschka) breaks down the architecture of the new Kimi K3 open model — attention design, MoE structure, and training notes. Useful if you're evaluating open-weight models for local or self-hosted inference.
Peter Norvig's classic 1998 essay arguing that mastery in programming takes roughly a decade of deliberate practice, not a quick bootcamp. Well-known, frequently reposted content with nothing new.
An open-source notetaking app explicitly designed to be used by both humans and AI agents as a shared memory/context store. Conceptually adjacent to agent-memory/second-brain systems — worth a look for ideas or as a potential integration, though it's an early-stage project.
Routine version release of SBCL, a Common Lisp implementation. Standard point release with no described major user-facing changes.
A magazine/publication promoting its 'slow journalism' approach of covering news well after the initial cycle. General media/publishing content, not software-relevant.
Duplicate submission of the same multi-seat Wayland article already covered from the Lobsters source under a different HN title.
Apple is replacing its iPhone Upgrade Program with a new 'Apple Upgrade' plan. Consumer pricing/product news with no relevance to software development or business tooling decisions.
A deep technical dive into how the Zig compiler implements incremental compilation internally. High-quality systems writing, but scoped to a language not part of a typical Node/Postgres/Python solo-dev stack.
A video interview with Boris Cherny, creator of Claude Code, likely covering the tool's design philosophy and roadmap direction. Worth watching for anyone building on or heavily using Claude Code / agent-based dev tooling, though it's a video with no transcript summary available yet.
Product launch for a repairable, USB-C, developer-friendly GPS smartwatch. Consumer hardware gadget, not directly relevant to software business decisions.
Opinion piece arguing that AI models should be granted access to the ACM's paywalled digital library of computer science research. Editorial/advocacy piece, not a reproducible or actionable development.
Anthropic published research demonstrating Claude being used to discover real cryptographic weaknesses/vulnerabilities. A concrete, vendor-verified example of using an AI agent for practical security research — directly useful if you want to apply the same approach to auditing your own projects (e.g. Veremun/ContractorVerify's compliance-sensitive code).
OpenAI field report describing how scientists use AI coding agents to modernize scientific-computing workflows in genomics and related fields. Vendor-published case study scoped to a specific research niche (genomics), not directly applicable to a solo software builder's own agent tooling.
OpenAI status-page incident report: elevated 'invalid_prompt' error rates on Chat Completions, now fully resolved. Minor, already-resolved infrastructure incident with no lasting relevance.
Original markdown
# Nightly Librarian — Newsletter draft
Run: 9c5e6629-b756-45cd-a756-e8b827dccb96
Started: 2026-07-30T06:09:29.005Z
Completed: 2026-07-30T06:22:20.534Z
## Worth attention
- **Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident**
https://huggingface.co/blog/agent-intrusion-technical-timeline
A detailed technical postmortem of a real July 2026 security incident involving an intrusion via a frontier AI lab's coding agent, including the attack timeline and how it was detected. Directly relevant to anyone running autonomous coding/agent workflows with broad tool access — worth reading to check your own agent permission boundaries and logging.
- **Codex Security**
https://github.com/openai/codex-security
OpenAI has published a dedicated security repo/guidance for its Codex coding agent, covering sandboxing, permission boundaries, and safe-usage practices. Directly relevant if you run Codex (or similar coding agents) with filesystem/shell access — worth reviewing your own agent permission setup against it.
- **Discovering Cryptographic Weaknesses with Claude**
https://www.anthropic.com/research/discovering-cryptographic-weaknesses
Anthropic published research demonstrating Claude being used to discover real cryptographic weaknesses/vulnerabilities. A concrete, vendor-verified example of using an AI agent for practical security research — directly useful if you want to apply the same approach to auditing your own projects (e.g. Veremun/ContractorVerify's compliance-sensitive code).
- **Building (systems) software with Nix**
https://hondu.co/blog/building-systems-software
Practical write-up on using Nix to build and package lower-level systems software, covering pain points and patterns for reproducible builds. Useful reference for anyone considering Nix for deployment reproducibility, though it's a narrow tooling choice.
- **Kimi K3 Architecture Overview and Notes**
https://sebastianraschka.com/blog/2026/kimi-k3-architecture-notes.html
A well-regarded ML educator (Sebastian Raschka) breaks down the architecture of the new Kimi K3 open model — attention design, MoE structure, and training notes. Useful if you're evaluating open-weight models for local or self-hosted inference.
- **Parallel JSON parsing on the GPU with compute shaders**
https://github.com/friendlymatthew/slurpjson#slurpjson
Open-source project (slurpjson) demonstrating GPU-accelerated JSON parsing using compute shaders, with working code on GitHub. A niche but concrete performance-engineering technique worth knowing about if JSON parsing throughput ever becomes a bottleneck.
- **You Could Have Come Up With Kimi Delta Attention**
https://blog.doubleword.ai/you-could-have-come-up-with-kimi-delta-attention
An accessible technical explainer walking through the intuition behind Kimi's 'delta attention' mechanism, aimed at building understanding of a newer attention-architecture variant. Useful background if you're evaluating or fine-tuning models that use this approach, but educational rather than immediately actionable.
- **Substack writers, you need a website**
https://elizabethtai.com/2026/06/10/substack-writers-you-need-a-website/
Argues that writers relying solely on Substack should also own a personal website/domain for control, SEO, and portability, with practical reasoning about platform risk. Relevant to anyone running a public-facing newsletter or content site, including your own librarian digest site.
- **Hubble: Open-source notetaking app for you and your agents**
https://www.hubble.md/
An open-source notetaking app explicitly designed to be used by both humans and AI agents as a shared memory/context store. Conceptually adjacent to agent-memory/second-brain systems — worth a look for ideas or as a potential integration, though it's an early-stage project.
- **Interview with Boris Cherny [video]**
https://www.youtube.com/watch?v=qyPCVqFUyDo
A video interview with Boris Cherny, creator of Claude Code, likely covering the tool's design philosophy and roadmap direction. Worth watching for anyone building on or heavily using Claude Code / agent-based dev tooling, though it's a video with no transcript summary available yet.
## Full digest
- [R] [lobsters] Why Rocq is better than Lean for program verification — https://joomy.korkutblech.com/posts/2026-07-28-why-rocq-is-better.html — Personal essay arguing Rocq (formerly Coq) remains preferable to Lean for formal program verification work, pushing back against the growing popularity of Lean. Niche academic tooling debate with no bearing on typical solo-builder workflows.
- [R] [lobsters] The mean means nothing — https://fzakaria.com/2026/07/27/the-mean-means-nothing — Statistics-focused post arguing the arithmetic mean is frequently a misleading summary statistic, with examples of when medians or other measures serve better. General audience math writing, not software-specific.
- [R] [lobsters] State of multi-player Wayland — https://blinry.org/multi-seat-wayland/ — Technical deep-dive on multi-seat Wayland setups (multiple users/cursors on one Linux machine), covering current tooling gaps and workarounds. Interesting Linux desktop internals but a narrow hobbyist use case.
- [R] [lobsters] Manganin: tools matter — https://blog.manganin.dev/blog/tools-matter/ — Blog post reflecting on how the choice of developer tools shapes what an engineer builds. General musing without concrete recommendations or data.
- [R] [lobsters] Writing Toy Software Is A Joy (2025) — https://blog.jsbarretto.com/post/software-is-joy — Personal essay celebrating the fun of building small, low-stakes 'toy' software projects outside of professional pressure. Reposted 2025 piece with no new information.
- [P] [lobsters] Building (systems) software with Nix — https://hondu.co/blog/building-systems-software — Practical write-up on using Nix to build and package lower-level systems software, covering pain points and patterns for reproducible builds. Useful reference for anyone considering Nix for deployment reproducibility, though it's a narrow tooling choice.
- [R] [lobsters] What even are microservices? — https://var0.xyz/posts/what-even-are-microservices.html — Essay questioning the fuzzy definition of 'microservices' and how the term gets used inconsistently across the industry. General architecture commentary without new data.
- [R] [lobsters] Dependency Cultures - Richard Feldman (Software Should Work Conf 2026) — https://www.youtube.com/watch?v=E82ly38YEEQ — Conference talk from Richard Feldman on how different programming communities develop different norms around dependency usage. Thoughtful but a video talk with no transcript/summary provided in the feed.
- [P] [lobsters] Parallel JSON parsing on the GPU with compute shaders — https://github.com/friendlymatthew/slurpjson#slurpjson — Open-source project (slurpjson) demonstrating GPU-accelerated JSON parsing using compute shaders, with working code on GitHub. A niche but concrete performance-engineering technique worth knowing about if JSON parsing throughput ever becomes a bottleneck.
- [P] [lobsters] Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident — https://huggingface.co/blog/agent-intrusion-technical-timeline — A detailed technical postmortem of a real July 2026 security incident involving an intrusion via a frontier AI lab's coding agent, including the attack timeline and how it was detected. Directly relevant to anyone running autonomous coding/agent workflows with broad tool access — worth reading to check your own agent permission boundaries and logging.
- [R] [lobsters] Where are the third places for learners? — https://kernelkennel.com/blog/where-is-the-middle-space/ — Essay on the lack of informal community spaces ('third places') for people learning to code or other skills outside formal institutions. General social commentary, not software-specific.
- [P] [lobsters] You Could Have Come Up With Kimi Delta Attention — https://blog.doubleword.ai/you-could-have-come-up-with-kimi-delta-attention — An accessible technical explainer walking through the intuition behind Kimi's 'delta attention' mechanism, aimed at building understanding of a newer attention-architecture variant. Useful background if you're evaluating or fine-tuning models that use this approach, but educational rather than immediately actionable.
- [R] [lobsters] Starling, a Linux Desktop written from scratch — https://starling.build — A hobby project building a Linux desktop environment entirely from scratch. Interesting systems-programming showcase but an early-stage hobby OS project with no user-facing maturity yet.
- [R] [lobsters] Lobste.rs on Spinel — https://intertwingly.net/blog/2026/07/28/What-Happens-After-You-IPO.html — Meta post about the Lobsters community site's hosting/infrastructure situation, referencing a piece on 'what happens after you IPO.' Site-operations trivia with no external relevance.
- [R] [lobsters] The Cipher Behind QSYRUPWD: Reconstructing IBM i Password Hashes — https://blog.silentsignal.eu/2026/07/28/the-cipher-behind-qsyrupwd-reconstructing-ibm-i-password-hashes/ — Deep technical reverse-engineering writeup on how IBM i (AS/400) stores and hashes passwords via the undocumented QSYRUPWD mechanism. Rigorous security research, but scoped to a legacy enterprise platform far outside a typical solo builder's stack.
- [R] [lobsters] CP/M-68K emulator — https://kobolt.github.io/article-264.html — A hobby emulator project for the retro CP/M-68K operating system. Fun retrocomputing work, purely a nostalgia/hobby item.
- [R] [lobsters] Proof types in Dart: Using final classes as computational witnesses — https://modulovalue.com/blog/proof-types-in-dart/ — A technical post exploring how to encode 'proof types' (computational witnesses) using final classes in Dart's type system. Niche language-design technique for a language most solo builders in infra/backend work don't use.
- [R] [hn-top] More Tailscale tricks for your jailbroken Kindle — https://tailscale.com/blog/jailbroken-kindle-proxy-tun-modes — A fun hack showing how to run Tailscale on a jailbroken Kindle to proxy traffic. Entertaining hobbyist networking trick, not a work-relevant technique.
- [R] [hn-top] User Interfaces of the Demo Scene — https://www.datagubbe.se/scenegui/ — A retrospective look at UI design in the demoscene (real-time graphics/audio art community). Interesting design history but pure entertainment/nostalgia content.
- [P] [hn-top] Codex Security — https://github.com/openai/codex-security — OpenAI has published a dedicated security repo/guidance for its Codex coding agent, covering sandboxing, permission boundaries, and safe-usage practices. Directly relevant if you run Codex (or similar coding agents) with filesystem/shell access — worth reviewing your own agent permission setup against it.
- [R] [hn-top] Show HN: I was tired of opening 2 tabs for every HN link, so I made a userscript — https://github.com/twalichiewicz/HNewhere — A personal userscript that merges an HN-linked article and its comments discussion into one resizable side-panel view. Handy personal productivity tool but very narrow scope.
- [R] [hn-top] LearnVector – Andrew Ng's AI company building one‑to‑one learning experiences — https://learnvector.ai/ — Launch announcement for LearnVector, a new AI education company from Andrew Ng focused on one-to-one personalized learning. Notable founder, but a consumer ed-tech product with no direct relevance to a solo software builder's tooling or business decisions.
- [P] [hn-top] Substack writers, you need a website — https://elizabethtai.com/2026/06/10/substack-writers-you-need-a-website/ — Argues that writers relying solely on Substack should also own a personal website/domain for control, SEO, and portability, with practical reasoning about platform risk. Relevant to anyone running a public-facing newsletter or content site, including your own librarian digest site.
- [R] [hn-top] Half-Life ported to Mac OS 9 — https://mac-classic.com/news/half-life-ported-to-mac-os-9/ — A retro-computing project porting the 1998 game Half-Life to run natively on classic Mac OS 9. Impressive engineering feat but pure nostalgia/entertainment content.
- [R] [hn-top] ReFrame – The EPaper Camera — https://reframe.camera/ — Product launch for an e-paper display camera/photo device. Consumer hardware gadget, not software/dev-relevant.
- [R] [hn-top] Truth is not a direction: a Tarski attack on LLM probes — https://abeljansma.nl/2026/07/10/truth-is-not-a-direction.html — A theoretical critique arguing that 'linear probe' techniques for detecting truthfulness inside LLM activations are conceptually flawed, drawing on Tarski's semantic theory of truth. Interesting interpretability research but abstract and not tied to a reproducible tool or actionable change.
- [P] [hn-top] Kimi K3 Architecture Overview and Notes — https://sebastianraschka.com/blog/2026/kimi-k3-architecture-notes.html — A well-regarded ML educator (Sebastian Raschka) breaks down the architecture of the new Kimi K3 open model — attention design, MoE structure, and training notes. Useful if you're evaluating open-weight models for local or self-hosted inference.
- [R] [hn-top] Teach yourself programming in ten years (1998) — https://www.norvig.com/21-days.html — Peter Norvig's classic 1998 essay arguing that mastery in programming takes roughly a decade of deliberate practice, not a quick bootcamp. Well-known, frequently reposted content with nothing new.
- [P] [hn-top] Hubble: Open-source notetaking app for you and your agents — https://www.hubble.md/ — An open-source notetaking app explicitly designed to be used by both humans and AI agents as a shared memory/context store. Conceptually adjacent to agent-memory/second-brain systems — worth a look for ideas or as a potential integration, though it's an early-stage project.
- [R] [hn-top] Steel Bank Common Lisp version 2.6.7 — https://sbcl.org/all-news.html?2.6.7 — Routine version release of SBCL, a Common Lisp implementation. Standard point release with no described major user-facing changes.
- [R] [hn-top] Delayed Gratification – Proud to Be 'Last to Breaking News' — https://www.slow-journalism.com/ — A magazine/publication promoting its 'slow journalism' approach of covering news well after the initial cycle. General media/publishing content, not software-relevant.
- [R] [hn-top] Multiple Mouse Cursors in Wayland — https://blinry.org/multi-seat-wayland/ — Duplicate submission of the same multi-seat Wayland article already covered from the Lobsters source under a different HN title.
- [R] [hn-top] The iPhone Upgrade Program is being replaced by Apple Upgrade — https://www.apple.com/shop/iphone/iphone-upgrade-program — Apple is replacing its iPhone Upgrade Program with a new 'Apple Upgrade' plan. Consumer pricing/product news with no relevance to software development or business tooling decisions.
- [R] [hn-top] Zig's Incremental Compilation Internals — https://mlugg.co.uk/posts/incremental-compilation-internals/ — A deep technical dive into how the Zig compiler implements incremental compilation internally. High-quality systems writing, but scoped to a language not part of a typical Node/Postgres/Python solo-dev stack.
- [P] [hn-top] Interview with Boris Cherny [video] — https://www.youtube.com/watch?v=qyPCVqFUyDo — A video interview with Boris Cherny, creator of Claude Code, likely covering the tool's design philosophy and roadmap direction. Worth watching for anyone building on or heavily using Claude Code / agent-based dev tooling, though it's a video with no transcript summary available yet.
- [R] [hn-top] Una GPS smart watch – Repairable, USB-C charging, developer-friendly — https://unawatch.com/ — Product launch for a repairable, USB-C, developer-friendly GPS smartwatch. Consumer hardware gadget, not directly relevant to software business decisions.
- [R] [hn-top] Now is the time to give LLMs access to the ACM digital library — https://cacm.acm.org/opinion/now-is-the-time-to-give-llms-access-to-the-acm-digital-library/ — Opinion piece arguing that AI models should be granted access to the ACM's paywalled digital library of computer science research. Editorial/advocacy piece, not a reproducible or actionable development.
- [P] [hn-top] Discovering Cryptographic Weaknesses with Claude — https://www.anthropic.com/research/discovering-cryptographic-weaknesses — Anthropic published research demonstrating Claude being used to discover real cryptographic weaknesses/vulnerabilities. A concrete, vendor-verified example of using an AI agent for practical security research — directly useful if you want to apply the same approach to auditing your own projects (e.g. Veremun/ContractorVerify's compliance-sensitive code).
- [R] [openai-blog] Scientific computing in the age of agentic AI — https://openai.com/index/scientific-computing-agentic-ai — OpenAI field report describing how scientists use AI coding agents to modernize scientific-computing workflows in genomics and related fields. Vendor-published case study scoped to a specific research niche (genomics), not directly applicable to a solo software builder's own agent tooling.
- [R] [openai-status] Elevated error rates with the invalid_prompt error code — https://status.openai.com//incidents/01KYNNX2TQ3PV0KPJ9PTRRPX6C — OpenAI status-page incident report: elevated 'invalid_prompt' error rates on Chat Completions, now fully resolved. Minor, already-resolved infrastructure incident with no lasting relevance.