Table of Contents
Table of Contents
(daily refresh)
#
- Rejected the one new star, pfnet-research/meta-fuse-csi-plugin (a Kubernetes CSI storage plugin, out-of-scope infrastructure) [glm-5.3-flash]
- Section roots: model-selection added gpt-6.1-sol (2026-09-29, $2/$10 with cache halved) and remeasured the Kimi K3 floor to $0.38/$10.00 and k2.7-code to $0.67/$3.35; the landscape folded Orca and DeepSeek Harness [glm-5.3-flash]
- Harnesses: corrected pi’s “no MCP by design” claim (v0.99.0 shipped MCP as built-in extensions, v1.0.0 followed), recorded kimi-code’s K3 drop to $0.4255/$10.00, and folded release movement across the other members [glm-5.3-flash]
- Model access and assistant runtimes: recorded ChatGPT Pro’s $500/month third tier, Experiential’s credit ladder rework, and OpenWork’s eleventh pricing churn, each with a Price history row [glm-5.3-flash]
- Orchestration: recorded agno v3.1.0 (breaking), looptroop v0.6.0, mastra 1.74.0, orca v1.4.218, sim v0.9.9, superset v1.33.0, and others, and corrected the happy-coder star-ranking claim [glm-5.3-flash]
- Sandboxing and executions: recorded OpenShell’s fold into NVIDIA’s Open Agent Safety Platform (which explains its star surge), opensandbox’s wheel-fix prerelease, aigate’s move to dormancy, and the github-agentic-workflows prerelease train [glm-5.3-flash]
- Task management: recorded ordewell v0.5.6 and beads v1.3.1 with matrix cells moved [glm-5.3-flash]
- Memory, retrieval, and context engines: recorded LangChain’s move to LSU metering at $1.00 (Price history row) and releases for graft, graphify, knowhere, and cognee [glm-5.3-flash]
- Code review and evaluation: recorded ellipsis’s platform fee doubling to 20% of token cost and releases for langfuse, phoenix, and deepeval [glm-5.3-flash]
- Automated research: corrected AlphaProof’s API claim and Harmonic Aristotle’s funding and leadership ($120M Series C), and canonicalized openai-deep-research’s Wikipedia links [glm-5.3-flash]
- Control planes and skills: moved Veto and Code Atelier Governance SDK to dormant, folded Okto Neuron into the Okto Pulse note, and added missing model tags across seven notes [glm-5.3-flash]
- Repaired the protocols matrix, which had lagged its notes [glm-5.3-flash]
- Recorded ctx v2.2.5, kiro CLI 2.26.0, openchamber v2.1.0, spec-kit v1.0.13, and ouroboros v0.55.3, and moved super-simple-software-factory to quiet-since-launch [glm-5.3-flash]
- Rejected endstop.systems’ containment post and folded Andon’s benches into the Pion note [glm-5.3-flash]
- Queue: appended a question about rewording the section’s roughly 330 retained “real” uses [glm-5.3-flash]
- agents/_index.md: re-dated the matrix one-liners and updated several as-of lines; no membership changes [glm-5.3-flash]
(OpenRig orchestration note, owner-prompted)
#
- Added the openrig research note to Orchestration, where no existing article covered it [deepseek-v4.1-flash]
- openrig: recorded the Apache-2.0 self-hosted local control plane (rig CLI, terminal UI, and MCP server over tmux, turning Claude Code, Codex, and Pi sessions into a YAML-defined team of seats), the thin community footprint, the author’s in-thread isolation admission, and 3,873 stars with v0.6.4 [deepseek-v4.1-flash]
- orchestration-feature-matrix: extended thirty-nine to forty columns with OpenRig, every cell traced to the new note [deepseek-v4.1-flash]
- orchestration/_index.md gained the OpenRig entry; the section-index one-liner re-counted to forty [deepseek-v4.1-flash]
(Lean eval benchmark, owner-prompted)
#
- Added the Lean eval row to the model-benchmark-matrix Math group, moving the count to thirty-three [deepseek-v4.1-flash]
- Lean eval: recorded comparator-checked Lean formalization problems judged by the comparator and replayed through the nanoda kernel, submission-based ranks, and re-dated the intro to 2026-10-02 [deepseek-v4.1-flash]
- agents/_index.md: the model-benchmark-matrix one-liner re-counted to thirty-three [deepseek-v4.1-flash]