August 28, 2026
Report summary
10 stories cleared the bar, led by Changes to SourceHut's terms of service regarding LLMs, langchain-anthropic 1.7.0 adds Anthropic 1.0 SDK support and a skills container parameter, and tailcat: like netcat, but over Tailscale's data plane, without Tailscale's control plane.
Worth attention
SourceHut amended its terms of service to set explicit rules around LLM-generated content and LLM crawlers on the platform. Code-host policy changes are the category that quietly break assumptions about what you may host and how your code may be used. If you host or mirror anything on SourceHut, read the actual terms; if you are entirely on GitHub, note it as an early marker that forges are writing LLM policy into contract terms.
This release adds support for the Anthropic 1.0 SDK, a top-level container parameter for skills, and an updates thinking-display mode. If any agent code sits on langchain-anthropic, this is the release that moves you onto the 1.0 SDK line, and the skills container parameter is directly relevant to agent and MCP work. Read the diff before upgrading, because a major underlying SDK bump delivered in a minor version is where behaviour drifts quietly.
Tailscale released tailcat, a netcat-style tool that speaks its WireGuard-based data plane directly without requiring the coordination/control plane. For a one-person shop bridging a Mac and a VPS, this is a plausible route to authenticated point-to-point transport without standing up or depending on a full tailnet. Worth evaluating the next time you would reach for an SSH tunnel or ad-hoc port forward; treat it as new and unproven until you have run it yourself.
Gergely Orosz explains why the volume of PR and marketing podcast pitches he receives go unanswered, and what a pitch would need to look like to land. Narrow in scope, but concrete if you ever do outbound distribution for a product. No action unless podcast or media outreach is on your roadmap.
The langchain-core 1.6.1 release notes show a revert of the 1.6.2 release alongside a streaming-content indexing fix and a fix making StructuredTool JSON-serializable. The practical takeaway is narrow: do not pin 1.6.2. No action if you are not currently on the 1.6.x line.
The google-ai-changelog source emitted a Bengali-locale AI Studio link with no body content, whose URL references a gemini-3.7-flash model. Treat the model reference as an unconfirmed signal rather than an announcement, since no changelog text was captured. Separately, this indicates the google-ai-changelog scraper is picking up localized navigation junk and should be fixed.
An alpha release on the 1.4 line, including a fix for an MCP content type the conversion layer could not handle and test changes routing MCP tests through FastMCP. Relevant as a signal of where MCP support in langchain is heading, but alpha software should not go near anything you depend on.
This patch fixes PIIMiddleware so that redaction preserves content-block shape, plus a streaming content indexing fix in core. If you rely on langchain's PII redaction in a pipeline that handles user data, this is a correctness fix worth taking. Otherwise routine.
LAION published a large open video dataset. No description was captured in the feed, so scale, licensing and composition are unverified. Relevant only if video training data is on your roadmap.
The Rust project announced its first Maintainers in Residence, a funded arrangement for paid maintainer work on the ecosystem. No immediate action for a solo builder, but it is a data point on open source sustainability models that other large projects may copy.
Full digest
A founder building an HR SaaS asking for generic marketing help. No information content.
Open-ended request for startup advice after a previous product failed. No findings or data.
A discussion prompt about tooling for contact forms on small projects. No conclusion, benchmark or recommendation captured.
Self-promotion for a technology-lookup lead generation tool aimed at web design outreach. Promotional.
Idea validation request for a team management product. No findings.
Open-ended discussion prompt about pre-build research processes. No conclusions captured.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
The google-ai-changelog source emitted a Bengali-locale AI Studio link with no body content, whose URL references a gemini-3.7-flash model. Treat the model reference as an unconfirmed signal rather than an announcement, since no changelog text was captured. Separately, this indicates the google-ai-changelog scraper is picking up localized navigation junk and should be fixed.
Patch release whose only listed change is copying glibc into the distroless runners image as a matched pair. Build-plumbing fix with no user-facing behaviour change.
A moving beta tag carrying the identical 2.37.3 changelog. Duplicate of the 2.37.3 release item.
LAION published a large open video dataset. No description was captured in the feed, so scale, licensing and composition are unverified. Relevant only if video training data is on your roadmap.
A YC company launch post for an insurance brokerage serving robotics, drone, autonomous systems and satellite companies. Promotional launch content with no relevance to a solo software shop.
Obituary for the actor Tim Curry. Entertainment news, entirely out of scope for this digest.
An opinion post about using Obsidian with AI and the difficulty of following through on AI-suggested ideas. No body content was captured and there is no reproducible method or data behind it.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run.
Gergely Orosz explains why the volume of PR and marketing podcast pitches he receives go unanswered, and what a pitch would need to look like to land. Narrow in scope, but concrete if you ever do outbound distribution for a product. No action unless podcast or media outreach is on your roadmap.
SourceHut amended its terms of service to set explicit rules around LLM-generated content and LLM crawlers on the platform. Code-host policy changes are the category that quietly break assumptions about what you may host and how your code may be used. If you host or mirror anything on SourceHut, read the actual terms; if you are entirely on GitHub, note it as an early marker that forges are writing LLM policy into contract terms.
The Rust project announced its first Maintainers in Residence, a funded arrangement for paid maintainer work on the ecosystem. No immediate action for a solo builder, but it is a data point on open source sustainability models that other large projects may copy.
A 2002 essay on open source usability, resurfaced from the Internet Archive. Historically interesting but changes no decision today.
Tailscale released tailcat, a netcat-style tool that speaks its WireGuard-based data plane directly without requiring the coordination/control plane. For a one-person shop bridging a Mac and a VPS, this is a plausible route to authenticated point-to-point transport without standing up or depending on a full tailnet. Worth evaluating the next time you would reach for an SSH tunnel or ad-hoc port forward; treat it as new and unproven until you have run it yourself.
A video about how the Dolphin emulator handles a specific game's anti-piracy checks. Emulator trivia with no bearing on a solo software business.
Corporate announcement about OpenAI expanding market presence and developer engagement in Brazil. No product, pricing or API change.
An alpha release on the 1.4 line, including a fix for an MCP content type the conversion layer could not handle and test changes routing MCP tests through FastMCP. Relevant as a signal of where MCP support in langchain is heading, but alpha software should not go near anything you depend on.
This patch fixes PIIMiddleware so that redaction preserves content-block shape, plus a streaming content indexing fix in core. If you rely on langchain's PII redaction in a pipeline that handles user data, this is a correctness fix worth taking. Otherwise routine.
Patch release for the Fireworks provider package, dropping reasoning history blocks and refreshing model profile data. No user-facing change unless you use Fireworks through langchain.
The langchain-core 1.6.1 release notes show a revert of the 1.6.2 release alongside a streaming-content indexing fix and a fix making StructuredTool JSON-serializable. The practical takeaway is narrow: do not pin 1.6.2. No action if you are not currently on the 1.6.x line.
This release adds support for the Anthropic 1.0 SDK, a top-level container parameter for skills, and an updates thinking-display mode. If any agent code sits on langchain-anthropic, this is the release that moves you onto the 1.0 SDK line, and the skills container parameter is directly relevant to agent and MCP work. Read the diff before upgrading, because a major underlying SDK bump delivered in a minor version is where behaviour drifts quietly.
Original markdown
# Nightly Librarian — Newsletter draft Run: 45fdd1d9-ba54-4e3d-b355-975a200d6e27 Started: 2026-08-28T06:11:42.117Z Completed: 2026-08-28T06:17:02.515Z ## Worth attention - **Changes to SourceHut's terms of service regarding LLMs** https://sourcehut.org/blog/2026-08-27-tos-changes-and-llms/ SourceHut amended its terms of service to set explicit rules around LLM-generated content and LLM crawlers on the platform. Code-host policy changes are the category that quietly break assumptions about what you may host and how your code may be used. If you host or mirror anything on SourceHut, read the actual terms; if you are entirely on GitHub, note it as an early marker that forges are writing LLM policy into contract terms. - **langchain-anthropic 1.7.0 adds Anthropic 1.0 SDK support and a skills container parameter** https://github.com/langchain-ai/langchain/releases/tag/langchain-anthropic%3D%3D1.7.0 This release adds support for the Anthropic 1.0 SDK, a top-level container parameter for skills, and an updates thinking-display mode. If any agent code sits on langchain-anthropic, this is the release that moves you onto the 1.0 SDK line, and the skills container parameter is directly relevant to agent and MCP work. Read the diff before upgrading, because a major underlying SDK bump delivered in a minor version is where behaviour drifts quietly. - **tailcat: like netcat, but over Tailscale's data plane, without Tailscale's control plane** https://github.com/tailscale/tailcat Tailscale released tailcat, a netcat-style tool that speaks its WireGuard-based data plane directly without requiring the coordination/control plane. For a one-person shop bridging a Mac and a VPS, this is a plausible route to authenticated point-to-point transport without standing up or depending on a full tailnet. Worth evaluating the next time you would reach for an SSH tunnel or ad-hoc port forward; treat it as new and unproven until you have run it yourself. - **Why you're not getting a response to your podcast pitch from me (or others)** https://blog.pragmaticengineer.com/why-youre-not-getting-a-response-to-your-podcast-pitch-from-me-or-others/ Gergely Orosz explains why the volume of PR and marketing podcast pitches he receives go unanswered, and what a pitch would need to look like to land. Narrow in scope, but concrete if you ever do outbound distribution for a product. No action unless podcast or media outreach is on your roadmap. - **langchain-core 1.6.1 reverts the 1.6.2 release** https://github.com/langchain-ai/langchain/releases/tag/langchain-core%3D%3D1.6.1 The langchain-core 1.6.1 release notes show a revert of the 1.6.2 release alongside a streaming-content indexing fix and a fix making StructuredTool JSON-serializable. The practical takeaway is narrow: do not pin 1.6.2. No action if you are not currently on the 1.6.x line. - **Google AI Studio changelog item referencing gemini-3.7-flash (broken localized scrape)** https://aistudio.google.com/prompts/new_chat?model=gemini-3.7-flash&hl=bn The google-ai-changelog source emitted a Bengali-locale AI Studio link with no body content, whose URL references a gemini-3.7-flash model. Treat the model reference as an unconfirmed signal rather than an announcement, since no changelog text was captured. Separately, this indicates the google-ai-changelog scraper is picking up localized navigation junk and should be fixed. - **langchain 1.4.0a1 alpha with MCP content-type fix** https://github.com/langchain-ai/langchain/releases/tag/langchain%3D%3D1.4.0a1 An alpha release on the 1.4 line, including a fix for an MCP content type the conversion layer could not handle and test changes routing MCP tests through FastMCP. Relevant as a signal of where MCP support in langchain is heading, but alpha software should not go near anything you depend on. - **langchain 1.3.18 fixes PIIMiddleware redaction** https://github.com/langchain-ai/langchain/releases/tag/langchain%3D%3D1.3.18 This patch fixes PIIMiddleware so that redaction preserves content-block shape, plus a streaming content indexing fix in core. If you rely on langchain's PII redaction in a pipeline that handles user data, this is a correctness fix worth taking. Otherwise routine. - **LAION Big Video Dataset** https://projects.laion.ai/bvd/ LAION published a large open video dataset. No description was captured in the feed, so scale, licensing and composition are unverified. Relevant only if video training data is on your roadmap. - **Rust announces its first Maintainers in Residence** https://blog.rust-lang.org/2026/08/26/announcing-our-first-maintainers-in-residence/ The Rust project announced its first Maintainers in Residence, a funded arrangement for paid maintainer work on the ecosystem. No immediate action for a solo builder, but it is a data point on open source sustainability models that other large projects may copy. ## Full digest - [R] [reddit-saas] Can any one help? — https://www.reddit.com/r/SaaS/comments/1vzmtth/can_any_one_help/ — A founder building an HR SaaS asking for generic marketing help. No information content. - [R] [reddit-saas] Building new Saas, any advice — https://www.reddit.com/r/SaaS/comments/1vzmo5u/building_new_saas_any_advice/ — Open-ended request for startup advice after a previous product failed. No findings or data. - [R] [reddit-saas] How do you handle contact forms on small projects? — https://www.reddit.com/r/SaaS/comments/1vzmb6f/how_do_you_handle_contact_forms_on_small_projects/ — A discussion prompt about tooling for contact forms on small projects. No conclusion, benchmark or recommendation captured. - [R] [reddit-saas] I built a thing that'll let you find companies by their technologies — https://www.reddit.com/r/SaaS/comments/1vzn1nx/i_built_a_thing_thatll_let_you_find_companies_by/ — Self-promotion for a technology-lookup lead generation tool aimed at web design outreach. Promotional. - [R] [reddit-saas] Building a team management SaaS, what features would you actually find useful? — https://www.reddit.com/r/SaaS/comments/1vzn01y/building_a_team_management_saas_what_features/ — Idea validation request for a team management product. No findings. - [R] [reddit-saas] what does your saas research process look like before you build? — https://www.reddit.com/r/SaaS/comments/1vzmo27/what_does_your_saas_research_process_look_like/ — Open-ended discussion prompt about pre-build research processes. No conclusions captured. - [R] [reddit-saas] Are chargebacks a data problem as much as a payments problem? — https://www.reddit.com/r/SaaS/comments/1vzmbs5/are_chargebacks_a_data_problem_as_much_as_a/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Most SaaS founders waste months building something nobody wants. — https://www.reddit.com/r/SaaS/comments/1vzm782/most_saas_founders_waste_months_building/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Slate - shadcn-inspired blade ui kit for laravel and livewire built using tailwind + alpine.js — https://www.reddit.com/r/SaaS/comments/1vzm765/slate_shadcninspired_blade_ui_kit_for_laravel_and/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] My last SaaS failed. So I built a tool to stop myself from making the same mistake again. — https://www.reddit.com/r/SaaS/comments/1vzm6y6/my_last_saas_failed_so_i_built_a_tool_to_stop/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Not able to comprehend.. what is wrong with perplexity...? Is it happening with everyone or just me.. — https://www.reddit.com/r/SaaS/comments/1vzm29y/not_able_to_comprehend_what_is_wrong_with/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [M] [google-ai-changelog] Google AI Studio changelog item referencing gemini-3.7-flash (broken localized scrape) — https://aistudio.google.com/prompts/new_chat?model=gemini-3.7-flash&hl=bn — The google-ai-changelog source emitted a Bengali-locale AI Studio link with no body content, whose URL references a gemini-3.7-flash model. Treat the model reference as an unconfirmed signal rather than an announcement, since no changelog text was captured. Separately, this indicates the google-ai-changelog scraper is picking up localized navigation junk and should be fixed. - [R] [gh-n8n] [email protected] — https://github.com/n8n-io/n8n/releases/tag/n8n%402.37.3 — Patch release whose only listed change is copying glibc into the distroless runners image as a matched pair. Build-plumbing fix with no user-facing behaviour change. - [R] [gh-n8n] n8n beta tag — https://github.com/n8n-io/n8n/releases/tag/beta — A moving beta tag carrying the identical 2.37.3 changelog. Duplicate of the 2.37.3 release item. - [M] [hn-top] LAION Big Video Dataset — https://projects.laion.ai/bvd/ — LAION published a large open video dataset. No description was captured in the feed, so scale, licensing and composition are unverified. Relevant only if video training data is on your roadmap. - [R] [hn-top] Launch HN: Risklytics (YC S26) – Insurance brokerage for frontier tech companies — https://www.risklytics.ai/ — A YC company launch post for an insurance brokerage serving robotics, drone, autonomous systems and satellite companies. Promotional launch content with no relevance to a solo software shop. - [R] [hn-top] Tim Curry has died — https://www.theguardian.com/film/2026/aug/26/tim-curry-dies-rocky-horror-show-stephen-king-it-legend-film — Obituary for the actor Tim Curry. Entertainment news, entirely out of scope for this digest. - [R] [hn-top] It's so hard to finish an idea that is not yours and is just suggested by AI — https://www.ssp.sh/brain/using-obsidian-with-ai/ — An opinion post about using Obsidian with AI and the difficulty of following through on AI-suggested ideas. No body content was captured and there is no reproducible method or data behind it. - [R] [reddit-saas] First customer on my PaaS Startup Peon.sh — https://www.reddit.com/r/SaaS/comments/1vzo7vf/first_customer_on_my_paas_startup_peonsh/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] More original content, fewer AI posts — https://www.reddit.com/r/SaaS/comments/1vzo2bl/more_original_content_fewer_ai_posts/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Built a portfolio tracker with an MCP server so Claude and Cursor can use portfolios as context — https://www.reddit.com/r/SaaS/comments/1vzpay5/wanted_to_replace_yahoo_finance_as_my_portfolio/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] My SaaS-GEO playbook — https://www.reddit.com/r/SaaS/comments/1vzp61l/my_saasgeo_playbook/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] I built a resume feedback tool and need advice on growth — https://www.reddit.com/r/SaaS/comments/1vzop05/i_built_a_resume_feedback_tool_and_need_advice_on/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Thinking to build expert Developers/Solopreneurs group — https://www.reddit.com/r/SaaS/comments/1vzohm5/thinking_to_build_expert_developerssolopreneurs/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Is paying to increase DR/DA actually worth it? — https://www.reddit.com/r/SaaS/comments/1vzo4iw/is_paying_to_increase_drda_actually_worth_it/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Getting started is the hardest part in the marketing. Once you are there you have to swim you like it or not. — https://www.reddit.com/r/SaaS/comments/1vzqish/getting_started_is_the_hardest_part_in_the/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] Conversation as the key to product market fit — https://www.reddit.com/r/SaaS/comments/1vzqg4p/conversation_as_the_key_to_product_market_fit/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [R] [reddit-saas] How long do you need to sell B2B SaaS before you can conclude there is no product-market fit? — https://www.reddit.com/r/SaaS/comments/1vzqg0n/how_long_do_you_need_to_sell_b2b_saas_before_you/ — Deferred by pre-triage over the per-run item cap and not individually reviewed. Recorded as rejected for this run. - [P] [pragmatic-engineer] Why you're not getting a response to your podcast pitch from me (or others) — https://blog.pragmaticengineer.com/why-youre-not-getting-a-response-to-your-podcast-pitch-from-me-or-others/ — Gergely Orosz explains why the volume of PR and marketing podcast pitches he receives go unanswered, and what a pitch would need to look like to land. Narrow in scope, but concrete if you ever do outbound distribution for a product. No action unless podcast or media outreach is on your roadmap. - [P] [lobsters] Changes to SourceHut's terms of service regarding LLMs — https://sourcehut.org/blog/2026-08-27-tos-changes-and-llms/ — SourceHut amended its terms of service to set explicit rules around LLM-generated content and LLM crawlers on the platform. Code-host policy changes are the category that quietly break assumptions about what you may host and how your code may be used. If you host or mirror anything on SourceHut, read the actual terms; if you are entirely on GitHub, note it as an early marker that forges are writing LLM policy into contract terms. - [M] [lobsters] Rust announces its first Maintainers in Residence — https://blog.rust-lang.org/2026/08/26/announcing-our-first-maintainers-in-residence/ — The Rust project announced its first Maintainers in Residence, a funded arrangement for paid maintainer work on the ecosystem. No immediate action for a solo builder, but it is a data point on open source sustainability models that other large projects may copy. - [R] [lobsters] Why Free Software usability tends to suck (2002) — https://web.archive.org/web/20030201183139/http://mpt.phrasewise.com/discuss/msgReader$173 — A 2002 essay on open source usability, resurfaced from the Internet Archive. Historically interesting but changes no decision today. - [P] [lobsters] tailcat: like netcat, but over Tailscale's data plane, without Tailscale's control plane — https://github.com/tailscale/tailcat — Tailscale released tailcat, a netcat-style tool that speaks its WireGuard-based data plane directly without requiring the coordination/control plane. For a one-person shop bridging a Mac and a VPS, this is a plausible route to authenticated point-to-point transport without standing up or depending on a full tailnet. Worth evaluating the next time you would reach for an SSH tunnel or ad-hoc port forward; treat it as new and unproven until you have run it yourself. - [R] [lobsters] How Dolphin deals with the anti-piracy of "The Smurfs Dance Party" — https://www.youtube.com/watch?v=wN2WdO4c_jM — A video about how the Dolphin emulator handles a specific game's anti-piracy checks. Emulator trivia with no bearing on a solo software business. - [R] [openai-blog] Expanding OpenAI's presence in Brazil — https://openai.com/index/expanding-our-presence-in-brazil — Corporate announcement about OpenAI expanding market presence and developer engagement in Brazil. No product, pricing or API change. - [M] [gh-langchain] langchain 1.4.0a1 alpha with MCP content-type fix — https://github.com/langchain-ai/langchain/releases/tag/langchain%3D%3D1.4.0a1 — An alpha release on the 1.4 line, including a fix for an MCP content type the conversion layer could not handle and test changes routing MCP tests through FastMCP. Relevant as a signal of where MCP support in langchain is heading, but alpha software should not go near anything you depend on. - [M] [gh-langchain] langchain 1.3.18 fixes PIIMiddleware redaction — https://github.com/langchain-ai/langchain/releases/tag/langchain%3D%3D1.3.18 — This patch fixes PIIMiddleware so that redaction preserves content-block shape, plus a streaming content indexing fix in core. If you rely on langchain's PII redaction in a pipeline that handles user data, this is a correctness fix worth taking. Otherwise routine. - [R] [gh-langchain] langchain-fireworks==1.6.1 — https://github.com/langchain-ai/langchain/releases/tag/langchain-fireworks%3D%3D1.6.1 — Patch release for the Fireworks provider package, dropping reasoning history blocks and refreshing model profile data. No user-facing change unless you use Fireworks through langchain. - [M] [gh-langchain] langchain-core 1.6.1 reverts the 1.6.2 release — https://github.com/langchain-ai/langchain/releases/tag/langchain-core%3D%3D1.6.1 — The langchain-core 1.6.1 release notes show a revert of the 1.6.2 release alongside a streaming-content indexing fix and a fix making StructuredTool JSON-serializable. The practical takeaway is narrow: do not pin 1.6.2. No action if you are not currently on the 1.6.x line. - [P] [gh-langchain] langchain-anthropic 1.7.0 adds Anthropic 1.0 SDK support and a skills container parameter — https://github.com/langchain-ai/langchain/releases/tag/langchain-anthropic%3D%3D1.7.0 — This release adds support for the Anthropic 1.0 SDK, a top-level container parameter for skills, and an updates thinking-display mode. If any agent code sits on langchain-anthropic, this is the release that moves you onto the 1.0 SDK line, and the skills container parameter is directly relevant to agent and MCP work. Read the diff before upgrading, because a major underlying SDK bump delivered in a minor version is where behaviour drifts quietly.