Daily AI Implementation Scout Council, 2026-09-16

Today's ranked top 20

newbuild nowBoth runtimes32 / 35

Claude Developer Platform, on-demand conversation compaction in the Messages API repo

On September 14, 2026, the Claude Messages API added on-demand conversation compaction, in beta under the compact-2026-09-04 header. Sending a top-level compaction parameter returns a signed summary block covering the messages sent so far, kept recent turns stay word for word, and the compaction can run in the background. On models with preserved thinking, the thinking in kept turns can stay valid.

What it does for you: This is a native, Anthropic-signed alternative to what Axion's context-mode plugin does by hand today: shrinking a long agent session before it runs out of context. Because compaction runs on demand and in the background, a long Hermes or Axion run could compact mid-task instead of hitting a hard context wall, which is exactly the failure mode that broke the scout newsletter itself in August 2026.

In practice: A first-party fix for a problem YY has already paid for in dead runs, worth a real trial rather than a shelf item.

For: Both runtimes. It is a plain Claude API parameter, callable from Claude Code, from a raw SDK script, or from Hermes wherever it calls the Messages API directly.

Security5

Quality5

Auditability4

Useful to you5

Useful to community4

Buildable now5

Hermes4

Verdict: build now. Low integration cost, official and signed, and it targets the same context exhaustion failure this very skill was rebuilt to avoid.

Build #1 Claude Developer Platform, on-demand conversation compaction in the Messages API: use the ai-implementation-build-intake skill to build this safely. Source: https://docs.claude.com/en/release-notes/overview.md. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Shipped September 14, 2026, beta header compact-2026-09-04, one top-level compaction parameter on the Messages API.

test firstBoth runtimes29 / 35

Qdrant Edge, in-process local vector database repo

Qdrant shipped Qdrant Edge, an in-process version of Qdrant that shares the same internals, storage format, and points API as the server version but runs locally with no separate service. It is compatible with the server version and can read shard snapshots the server version created.

What it does for you: Axion and Hermes both currently reach for file based or ad hoc search rather than a real local vector store. Qdrant Edge is a genuine fit for a local knowledge base like the 2nd Brain vaults, since it needs no server process and the storage format matches the hosted product if YY ever wants to graduate to it.

In practice: A real gap filled cleanly, worth a small pilot against one brain vault before committing.

For: Both runtimes. It runs in-process with no server to stand up, so it drops into a Hermes local runtime or an Axion skill script equally well.

Security4

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes5

Verdict: test first. New product line with a compatible storage format is promising, but it has not been run against a real Axion or Hermes workload yet.

Build #2 Qdrant Edge, in-process local vector database: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/qdrant/qdrant/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Documented in the 2026 Qdrant changelog as in-process, sharing internals, storage format, and points API with server-version Qdrant, and able to read server-version shard snapshots.

test firstBoth runtimes27 / 35

Braintrust JavaScript SDK v3.29.0, eval auto-instrumentation and a Claude Agent SDK span fix repo

Braintrust's JavaScript SDK reached v3.29.0, where braintrust eval now applies auto-instrumentation so supported AI clients emit spans before eval files load. The prior v3.28.0 release fixed Claude Agent SDK subagent tool span parenting and corrected per-call token and cost metrics so query-phase usage stopped being attributed to the final call.

What it does for you: The subagent span parenting fix is squarely about Claude Agent SDK subagents, which is exactly the orchestration pattern Axion and its Skills fleet run on. If YY ever wires up real observability instead of reading raw transcripts, this is the SDK version where subagent traces stop lying about which call cost what.

In practice: A quiet but correct fix to the exact shape of tool that would matter if observability ever gets built.

For: Both runtimes. The SDK is a plain npm package, usable from any Node or Claude Code host and equally from a Hermes-side Node process.

Security4

Quality4

Auditability4

Useful to you4

Useful to community4

Buildable now4

Hermes3

Verdict: test first. No current Axion or Hermes workflow runs Braintrust, but the subagent span fix is directly relevant if observability gets prioritized, worth a small pilot rather than a pass.

Build #3 Braintrust JavaScript SDK v3.29.0, eval auto-instrumentation and a Claude Agent SDK span fix: use the ai-implementation-build-intake skill to build this safely. Source: https://www.braintrust.dev/docs/changelog. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v3.29.0 adds eval auto-instrumentation, v3.28.0 fixed Claude Agent SDK subagent tool span parenting and per-call cost attribution.

test firstBoth runtimes24 / 35

Mem0 TypeScript SDK v3.1.0, 26 new providers and reranking repo

Mem0's TypeScript OSS SDK reached v3.1.0, its largest provider release to date, adding 17 new vector stores including Pinecone, Weaviate, Milvus, and Chroma, 5 new LLM providers including AWS Bedrock and xAI Grok, 4 new embedding providers, and 4 rerankers. The Python SDK shipped v2.0.12 alongside it with fixes and security patches.

What it does for you: This closes most of the gap between Mem0's TypeScript and Python SDKs and adds real provider breadth, but YY's own memory-tool-decision review already concluded no memory tool is needed after an 18 of 20 recall test passed without one. This stays a capability upgrade to watch, not a reason to revisit that call absent a new recall gap.

In practice: Genuinely large release, filed low on usefulness only because the prior decision already settled the question.

For: Both runtimes. Both the TypeScript and Python SDKs are plain packages, callable from Claude Code tooling or a Hermes process without a host-specific integration.

Security4

Quality4

Auditability3

Useful to you2

Useful to community4

Buildable now4

Hermes3

Verdict: test first. Strong engineering and breadth earn the score, but useful_yy is deliberately capped by the standing memory-tool-decision, so this is tracked rather than adopted.

Build #4 Mem0 TypeScript SDK v3.1.0, 26 new providers and reranking: use the ai-implementation-build-intake skill to build this safely. Source: https://docs.mem0.ai/changelog. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · TypeScript SDK v3.1.0 adds 17 new vector stores, 5 new LLM providers, 4 new embedding providers, and 4 rerankers, Python SDK v2.0.12 ships alongside it.

watchBoth runtimes21 / 35

Stagehand server-v3 v3.7.6, two patch releases past the tracked stable repo

Stagehand's server-v3 line moved from the v3.7.3 stable release tracked in the prior issue to v3.7.5 and then v3.7.6, both routine patch releases on the same server-v3 branch.

What it does for you: No current Axion or Hermes workflow runs Stagehand for browser automation, YY's browser work currently goes through Chrome MCP instead. This is tracked only to keep the version current if that changes.

In practice: Routine patch cadence, nothing structurally new to act on.

For: Both runtimes. Stagehand runs as a standalone browser automation server, callable from either a Claude Code skill or a Hermes process over its own API.

Security4

Quality4

Auditability3

Useful to you2

Useful to community3

Buildable now3

Hermes2

Verdict: watch. Two patch releases with no described behavior change and no current Axion browser-automation workflow using Stagehand.

Build #5 Stagehand server-v3 v3.7.6, two patch releases past the tracked stable: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/browserbase/stagehand/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Server-v3 branch moved from v3.7.3 to v3.7.5 then v3.7.6.

newwatchBoth runtimes20 / 35

Weaviate v1.39.4, two CVE dependency fixes repo

Weaviate reached v1.39.4, pinning a minimum OpenSSL version so cached Docker builds pull CVE-2026-46600 and CVE-2026-56852 fixes, alongside a fix for named-vector source properties being ignored on re-vectorize and a fix for shard store teardown racing in-flight references.

What it does for you: No current Axion or Hermes component runs Weaviate. The two CVE fixes are worth knowing about only if that changes, so this is a version bump to note, not a build to plan.

In practice: Security-relevant fix on infrastructure nobody here is running yet.

For: Both runtimes. Weaviate runs as a standalone database service, reachable from Claude Code or Hermes over its client libraries with no host-specific binding.

Security5

Quality4

Auditability3

Useful to you1

Useful to community3

Buildable now2

Hermes2

Verdict: watch. Two real CVE fixes justify a solid security score, but no current YY workflow runs Weaviate, so it is tracked rather than acted on.

Build #6 Weaviate v1.39.4, two CVE dependency fixes: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/weaviate/weaviate/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v1.39.4 pins OpenSSL to close CVE-2026-46600 and CVE-2026-56852, up from the 1.37.3 baseline tracked previously.

watchBoth runtimes20 / 35

MCP publishes its new roadmap, agentic messaging and agent identity repo

The Model Context Protocol team published a new roadmap naming agentic messaging primitives, HTTP-native transport unification and hardening, and agent identity and entitlements as priority areas, building on the 2026-07-28 specification that made MCP stateless at the protocol layer.

What it does for you: The Axion Telegram bridge and every MCP server YY runs sit directly on top of this protocol. Agent identity and entitlements in particular would matter to the ghl-write-gate and secrets-gate style hooks YY already relies on, since those are exactly the kind of per-agent permission boundary the roadmap is naming as a gap.

In practice: Forward planning worth logging, nothing to build until it ships.

For: Both runtimes. A protocol roadmap is not runtime-bound, it shapes every MCP server and client on either host equally.

Security3

Quality4

Auditability3

Useful to you3

Useful to community4

Buildable now1

Hermes2

Verdict: watch. A roadmap post names real gaps relevant to Axion's MCP usage, but there is no shipped feature yet to build against.

Build #7 MCP publishes its new roadmap, agentic messaging and agent identity: use the ai-implementation-build-intake skill to build this safely. Source: https://blog.modelcontextprotocol.io/posts/mcp-roadmap/. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Roadmap builds on the 2026-07-28 specification that introduced a stateless protocol core, Extensions, Tasks, and a formal deprecation policy.

watchClaude (Axion)19 / 35

Claude Code v2.1.270, git permission regression fix repo

Claude Code v2.1.270, released September 12, 2026, fixed read-only git commands in Bash unexpectedly asking for permission after a session had been running for a while, a regression introduced in v2.1.269.

What it does for you: YY runs Claude Code every day, so this closes a real annoyance where long sessions started re-prompting for permission on plain read-only git commands. It ships through the normal auto-update path, there is nothing to build or wire in.

In practice: A small daily-use fix, worth knowing landed, nothing to act on beyond that.

For: Claude (Axion). Claude Code specific CLI, the fix has no meaning outside that host.

Security4

Quality4

Auditability3

Useful to you2

Useful to community3

Buildable now2

Hermes1

Verdict: watch. Auto-applied bug fix to the CLI YY already runs, useful to know about but there is no build or test action to take.

Build #8 Claude Code v2.1.270, git permission regression fix: use the ai-implementation-build-intake skill to build this safely. Source: https://code.claude.com/docs/en/changelog. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v2.1.270 shipped September 12, 2026, fixing a regression from v2.1.269 that made read-only git commands prompt for permission in long sessions.

watchBoth runtimes18 / 35

LiteLLM, stable tag three minor versions behind the 1.100.1 nightly line repo

A version tracker report dated September 14, 2026 flags LiteLLM's stable channel pinned at 1.83.14-stable.patch.3 while the latest nightly tag is 1.100.1, a large numeric gap the tracker recommends holding on until the next .stable.patch.X tag lands.

What it does for you: LiteLLM is a common gateway for routing multiple model providers through one API, relevant if Axion or Hermes ever needs a single proxy in front of several LLM vendors. The gap itself is the useful fact, a reminder not to pull latest blind if LiteLLM is ever adopted as a gateway.

In practice: A version-hygiene warning more than a feature to chase.

For: Both runtimes. LiteLLM is a plain proxy or Python library, usable as a gateway in front of either a Claude Code skill or a Hermes process.

Security3

Quality3

Auditability3

Useful to you2

Useful to community3

Buildable now2

Hermes2

Verdict: watch. No current YY workflow runs LiteLLM, the item is filed to record the stable-versus-nightly gap in case that changes.

Build #9 LiteLLM, stable tag three minor versions behind the 1.100.1 nightly line: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/BerriAI/litellm/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Stable channel pinned at 1.83.14-stable.patch.3 against a latest nightly tag of 1.100.1, per the September 14, 2026 tracker report.

#10

watchBoth runtimes17 / 35

Docling, an OCR regression holds pipelines on v1.16.1 while tags reach v2.127.0 repo

Docling's latest tagged release is v2.127.0, but a version tracker report dated September 14, 2026 documents production installs still held on v1.16.1 because v1.17 introduced a RapidOCR ONNX regression that causes a startup failure without a pre-warm step.

What it does for you: YY has no current docling pipeline for document ingestion, but the pattern is a real cautionary note for any future PDF or document parsing build: a known regression sat unresolved across dozens of tag bumps, and jumping straight to latest would have hit it.

In practice: Useful mainly as a warning label for whenever document parsing gets built.

For: Both runtimes. Docling is a plain Python document-parsing library, importable from a Claude Code skill or a Hermes ingestion script alike.

Security3

Quality2

Auditability3

Useful to you2

Useful to community3

Buildable now2

Hermes2

Verdict: watch. No current Axion ingestion pipeline uses docling, but the unresolved OCR regression is worth remembering before ever pulling latest.

Build #10 Docling, an OCR regression holds pipelines on v1.16.1 while tags reach v2.127.0: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/docling-project/docling/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Latest tag v2.127.0 against installs held at v1.16.1 since the v1.17 RapidOCR ONNX regression, per the September 14, 2026 tracker report.

#11

newwatchBoth runtimes17 / 35

Open Notebook v1.14.0, jump from a 1.8.5 baseline repo

A version tracker report dated September 14, 2026 lists Open Notebook moving from a tracked 1.8.5 baseline to v1.14.0, a minor-line jump across six tagged versions.

What it does for you: Open Notebook is an open-source research and notebook tool in the same space as YY's 2nd Brain vaults, useful mainly as a comparison point rather than a replacement, since the brain vaults are Obsidian-native and already working.

In practice: A large version jump worth a glance, not a priority build.

For: Both runtimes. Open Notebook is a self-hosted app, reachable the same way from a Claude Code skill or a Hermes process calling its API.

Security3

Quality3

Auditability3

Useful to you2

Useful to community2

Buildable now2

Hermes2

Verdict: watch. No gap exists that this tool would close given the working Obsidian brain setup, tracked only for awareness.

Build #11 Open Notebook v1.14.0, jump from a 1.8.5 baseline: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/botAGI/AGmind/issues/193. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Version tracked moving from 1.8.5 to v1.14.0 per the September 14, 2026 report.

#12

newwatchStandalone tool16 / 35

AWS Bedrock AgentCore, Consent Portal for Identity repo

Amazon Bedrock AgentCore Identity added a Consent Portal in September 2026, a hosted page where end users grant an agent consent to access resources on their behalf. It requires an AgentCore Gateway configured with JWT inbound authentication and an identity provider whose permitted scopes include openid.

What it does for you: YY runs no AWS Bedrock AgentCore infrastructure today, so this has no direct build path, but the consent-before-access pattern is the same shape of problem the GHL write gate and secrets gate hooks already solve locally.

In practice: A real feature in a stack YY does not run, filed for the pattern rather than the product.

For: Standalone tool. This is an AWS managed service feature, used through the AWS console and SDK rather than embedded in either Claude Code or Hermes.

Security4

Quality4

Auditability2

Useful to you1

Useful to community3

Buildable now1

Hermes1

Verdict: watch. No AWS Bedrock AgentCore usage exists at Axion, and the closed AWS documentation gives no code to audit.

Build #12 AWS Bedrock AgentCore, Consent Portal for Identity: use the ai-implementation-build-intake skill to build this safely. Source: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/identity-consent-portal.html. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Announced in the September 2026 AgentCore release notes; Consent Portal requires an AgentCore Gateway with JWT inbound authentication and an identity provider with openid in its permitted scopes.

#13

newwatchStandalone tool16 / 35

AWS Bedrock AgentCore Evaluations, TypeScript framework support repo

Amazon Bedrock AgentCore Evaluations added TypeScript framework support in September 2026, evaluating agents built with the TypeScript versions of Strands Agents, LangGraph, OpenAI Agents, and the Vercel AI SDK alongside the existing Python support.

What it does for you: No Axion or Hermes workflow runs on AWS Bedrock AgentCore, so this stays a watch item, useful only if YY ever standardizes agent evaluation on AWS infrastructure.

In practice: An incremental widening of an AWS product's coverage, not a decision point today.

For: Standalone tool. An AWS managed evaluation service, configured through AWS rather than run inside either Claude Code or Hermes.

Security4

Quality4

Auditability2

Useful to you1

Useful to community3

Buildable now1

Hermes1

Verdict: watch. No current AWS AgentCore usage and no plan to adopt it, tracked for completeness of the AWS agent-tooling landscape.

Build #13 AWS Bedrock AgentCore Evaluations, TypeScript framework support: use the ai-implementation-build-intake skill to build this safely. Source: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/supported-frameworks.html. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Announced in the September 2026 AgentCore release notes: TypeScript support added for 4 frameworks, Strands Agents, LangGraph, OpenAI Agents, and the Vercel AI SDK, alongside prior Python framework support.

#14

newwatchBoth runtimes15 / 35

Letta v0.16.5, patch bump repo

Letta bumped from v0.16.4 to v0.16.5, a version-only chore commit with no functional changes listed in the release notes.

What it does for you: Letta's headline feature is sleep-time compute, a background agent that reorganizes memory while the primary agent idles, but YY's memory-tool-decision review already found no recall gap that any memory tool needs to close.

In practice: A thin release on a tool that is not currently a fit regardless of version.

For: Both runtimes. Letta runs as a standalone memory-agent server, reachable equally from a Claude Code skill or a Hermes process over its API.

Security3

Quality3

Auditability3

Useful to you1

Useful to community2

Buildable now2

Hermes1

Verdict: watch. Chore-only version bump, and the standing memory-tool-decision keeps useful_yy low independent of release quality.

Build #14 Letta v0.16.5, patch bump: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/letta-ai/letta/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v0.16.5 released as a version-only bump from v0.16.4, tracked in pull request 3202.

#15

watchBoth runtimes15 / 35

Qdrant Web UI, resharding controls and optimization visualization repo

The Qdrant Web UI added a detailed visualization of optimization progress, buttons for resharding control, a create-collection dialog that previews the exact command, and a restored and improved points filter and search bar, across pull requests 345, 341, 334, and 344.

What it does for you: No current Axion or Hermes component runs Qdrant's server product, so this is UI polish on infrastructure not yet in use here, tracked in case the standalone Qdrant Edge item above ever grows into a server deployment.

In practice: Solid dashboard polish, not itself a reason to adopt Qdrant.

For: Both runtimes. A web dashboard bundled with the Qdrant server, reachable from a browser regardless of which host is driving the agent.

Security3

Quality3

Auditability3

Useful to you1

Useful to community2

Buildable now2

Hermes1

Verdict: watch. UI improvements to a database YY does not currently run, filed for completeness alongside the Qdrant Edge item.

Build #15 Qdrant Web UI, resharding controls and optimization visualization: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/qdrant/qdrant/releases. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Four pull requests, 345, 341, 334, and 344, adding resharding buttons, an optimization progress view, a command-preview create dialog, and a restored points filter bar.

#16

watchBoth runtimes14 / 35

Milvus etcd component, v3.5.25 to v3.7.1 repo

A version tracker report dated September 14, 2026 lists the etcd component bundled with a Milvus deployment moving from v3.5.25 to v3.7.1, a minor version bump.

What it does for you: No current Axion or Hermes component runs Milvus, so this is a component-level bump inside infrastructure not in use here, tracked for RAG-lens completeness only.

In practice: A minor internal-dependency bump, low signal on its own.

For: Both runtimes. A backing component of a self-hosted Milvus deployment, host-agnostic by nature.

Security2

Quality2

Auditability3

Useful to you1

Useful to community2

Buildable now2

Hermes2

Verdict: watch. No current Milvus deployment at Axion, and this is a supporting component rather than Milvus itself.

Build #16 Milvus etcd component, v3.5.25 to v3.7.1: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/botAGI/AGmind/issues/193. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · etcd component version moved from v3.5.25 to v3.7.1 per the September 14, 2026 tracker report.

#17

newwatchStandalone tool14 / 35

Codex CLI v0.153.x, plugin CLI for remote marketplaces repo

OpenAI's Codex CLI reached v0.153.4 on September 4, 2026. Its plugin CLI can now list, install, and remove plugins from remote marketplaces, and Vim mode added undo and redo while preserving complete drafts including pasted content and attachments.

What it does for you: YY's daily coding agent is Claude Code, not Codex CLI, so this is competitive landscape awareness rather than a build. The plugin CLI pattern is the same shape of thing Claude Code's own plugin marketplace already does.

In practice: A competitor closing a feature gap, worth noting, not worth switching for.

For: Standalone tool. OpenAI's own coding CLI, a separate product from Claude Code and not something Hermes or Axion embeds.

Security3

Quality3

Auditability2

Useful to you1

Useful to community3

Buildable now1

Hermes1

Verdict: watch. Not YY's tool of choice, tracked for competitive awareness of the autonomous coding agent landscape.

Build #17 Codex CLI v0.153.x, plugin CLI for remote marketplaces: use the ai-implementation-build-intake skill to build this safely. Source: https://releasebot.io/updates/openai/codex. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · v0.153.4 released September 4, 2026, adding a remote-marketplace plugin CLI and Vim mode undo and redo.

#18

skipBoth runtimes8 / 35

RAGFlow Elasticsearch component, 9.4.0 to 9.5.3 repo

A version tracker report dated September 14, 2026 lists the Elasticsearch component bundled with a RAGFlow deployment moving from 9.4.0 to 9.5.3, a minor version bump with no described feature change.

What it does for you: No current Axion or Hermes component runs RAGFlow, and this is a minor dependency bump with no listed content, so it does not change anything YY would build.

In practice: Too thin to act on, logged only to keep the RAG lens honest about what it found.

For: Both runtimes. A backing search component of a self-hosted RAGFlow deployment, not tied to either host.

Security2

Quality2

Auditability2

Useful to you0

Useful to community1

Buildable now1

Hermes0

Verdict: skip. A dependency-only minor bump with no described change and no current RAGFlow usage at Axion.

Build #18 RAGFlow Elasticsearch component, 9.4.0 to 9.5.3: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/botAGI/AGmind/issues/193. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Elasticsearch component version moved from 9.4.0 to 9.5.3 per the September 14, 2026 tracker report.

#19

skipBoth runtimes9 / 35

SearXNG, 2026.5.10 to 2026.9.14 repo

A version tracker report dated September 14, 2026 lists SearXNG, a self-hosted private search engine sometimes used to feed agent web search, moving from a tracked 2026.5.10 build to 2026.9.14, a minor version bump with no described feature change.

What it does for you: YY's agent web search already goes through the built-in WebSearch tool, so a self-hosted metasearch engine solves a problem that is not currently open here.

In practice: No open gap for this to fill right now.

For: Both runtimes. A self-hosted metasearch engine, reachable over HTTP from either host without any host-specific binding.

Security2

Quality2

Auditability2

Useful to you0

Useful to community2

Buildable now1

Hermes0

Verdict: skip. No current search-infrastructure gap this would close, and the bump carries no described feature content.

Build #19 SearXNG, 2026.5.10 to 2026.9.14: use the ai-implementation-build-intake skill to build this safely. Source: https://github.com/botAGI/AGmind/issues/193. Save canonical skill/agent under AXION\Skills and AXION\Agents.

Source · Version moved from 2026.5.10 to 2026.9.14 per the September 14, 2026 tracker report.

#20

newskipClaude (Axion)8 / 35

claude-plugins-official, Claude for Small Business plugin added then removed the same day plugin

On September 15, 2026, anthropics/claude-plugins-official merged pull request 6145 adding a Claude for Small Business plugin, then merged pull request 6151 removing it the same day, both authored inside the same commit window.

What it does for you: There is nothing installable here, the plugin no longer exists in the marketplace as of the removal commit. It fills the plugin-scout lens for today, and its own removal is the finding: a same-day add-then-revert is a governance signal worth one line, not a plugin worth evaluating.

In practice: Churn, not a product, logged so the plugin lens does not go silent.

For: Claude (Axion). A Claude Code plugin, meaningful only inside that host's plugin marketplace.

Security3

Quality1

Auditability3

Useful to you0

Useful to community1

Buildable now0

Hermes0

Verdict: skip. The plugin was removed the same day it was added, there is nothing left to install or evaluate.

Install: not applicable, Anthropic added and then removed this plugin on September 15, 2026, nothing remains to install. Review before enabling if it ever reappears.

Source · Pull request 6145 added the plugin, pull request 6151 removed it, both merged on September 15, 2026.