Category / T-CAT / SOFTWARE
Software
Software / T-2026-4474
Pi's agent harness ships without a permission system, and that is the point
The Pi agent harness from earendil-works ships a unified LLM API, agent loop, TUI, and coding CLI, with no built-in permission system and a hardened npm supply chain.
Sourceearendil-works/pi
Software / T-2026-4984
llama.cpp 0.4.1 ships ternary MoE support and a JSON schema rewrite
llama.cpp v0.4.1 lands ternary MoE, three new architectures, and a GDN normalization fix affecting Qwen, Kimi, and GLM models.
Sourceggerganov/llama.cpp v0.4.1
Software / T-2026-7789
Agent Skills Wants to Be npm for AI Coding Agents. The 13.4% Problem Is Real.
Tech Leads Club's agent-skills registry scans every skill with Snyk and curates prompts by hand, betting the AI agent plugin market needs a trust layer more than a bigger catalog.
Sourcetech-leads-club/agent-skills
Software / T-2026-4106
Everything Claude Code: the 10-month battle for AI-native memory
WorldFlowAI's everything-claude-code repo reveals the real frontier of AI-assisted development: memory, not codegen.
SourceWorldFlowAI/everything-claude-code
Software / T-2026-8869
Magnitude's local inference server turns your agent into the installer
Magnitude profiles your machine, recommends local models, and wires them into the agent you already use.
Sourcemagnitudedev/magnitude
Software / T-2026-2995
gpt-6-astra in openai-python 3.8.0 signals a new frontier-model cadence
openai-python v3.8.0 adds gpt-6-astra support. A tiny changelog entry with big implications.
Sourceopenai/openai-python v3.8.0
Software / T-2026-0833
Humanizer's 35 tells: the open-source skill exposing AI writing's fingerprints
blader/humanizer turns Wikipedia's 35 'Signs of AI writing' into an agent skill. A sharp look at what it reveals.
Sourceblader/humanizer
Software / T-2026-5526
Vercel's portless kills localhost:3000, and that matters for AI agents
Vercel Labs ships portless, replacing port numbers with named .localhost URLs. The real story is what stable addresses mean for AI agents.
Sourcevercel-labs/portless
Software / T-2026-3870
Anthropic SDK v1.3.0 hints at enterprise agent governance layer
Anthropic's Python SDK v1.3.0 ships user profiles, compliance settings, and memory-store schema updates. A signal of enterprise agent governance.
Sourceanthropics/anthropic-sdk-python v1.3.0
Software / T-2026-3026
OpenAI's Python SDK v3.7.0: A Quiet Release With a Loud Signal
OpenAI's openai-python v3.7.0 ships a usage API overhaul and a numpy fix. The small release reveals bigger strategic signals.
Sourceopenai/openai-python v3.7.0
Software / T-2026-9319
CVE-2026-82542: A perfect 10.0 in a router nobody patches
Tenda HG10's formIPv6Routing buffer overflow scores a perfect CVSS 10.0. The real story is the unpatched IoT long tail.
SourceCVE-2026-82542 (CVSS 10 CRITICAL)
Software / T-2026-0423
MiniMind shows the 64M-parameter LLM is a viable teaching tool
A 64M-parameter model trained in 2 hours for $3 of GPU time makes the full LLM pipeline legible.
Sourcejingyaogong/minimind