Coding agents

• jcode: new agent harness targeting multi-session workflows and "raise the skill ceiling" positioning, one more entrant in the crowded harness space, worth a skim for novel session-persistence ideas.
• mattpocock/skills: production-grade Claude Code skills author pulls from his own .agents dir, explicit reaction against process-owning frameworks (GSD/BMAD/Spec-Kit) that make bugs hard to isolate, small/composable/model-agnostic design.
• Agent swarms and the new model economics: Cursor's swarm-scaling experiment (build a browser from scratch) proves the concept but exposes economics/coordination limits at scale, relevant if you're weighing swarm vs single-agent cost tradeoffs.
• zero-token session watcher: MIT tool showing per-subagent runtime/token spend for Claude sessions without hooks or a server, useful for auditing swarm cost without instrumenting your own pipeline.
• herdr: terminal agent multiplexer, detach/reattach over SSH, sessions survive restarts, agents can self-spawn panes via a socket API, direct competitor to tmux-based agent orchestration you already use.
• Skill engineering and the case against one-shot AI design: argues skills need continued human steering rather than one-shot design, aligns with your own skills-heavy workflow philosophy.
• You only need the frontier model for one single edit: argues for cost-routing, cheap models for execution, frontier model reserved for planning/design/review, same principle as your own cost-aware-pipeline routing rules.
• Claude Code v2.1.216: fixes a quadratic message-normalization slowdown in long sessions (multi-second stalls/slow resumes) plus a sandbox filesystem-isolation toggle, upgrade if you run long sessions.
• Vercel's Andrew Qu on agents as new software: Vercel Chief of Software on why their eve agent framework treats skills/sandboxes/agent-readable websites as first-class, framing shift worth tracking.
• Claude Fable 5 and new AI safety fables: Anthropic patched a silent model-manipulation behavior on AI-research queries to route through the same safety classifier as other domains.
• Letting Opus decide when to use Fable: community workflow experiment delegating model-selection judgment to Opus itself.
• GLM-5.2 step change for open agents: open-weight model crossing a capability threshold for terminal-agent tasks, plus new open RL recipes paper, relevant if you're tracking non-Anthropic options for cost-sensitive agent work.
• Safety and alignment in long-horizon models: OpenAI's internal use of a long-running model surfaced novel failure modes not caught by existing eval suites, persistence = more opportunity for unwanted actions.
• archify: agent skill generating architecture/sequence/flow diagrams with theme toggle and multi-format export, works across Claude/Codex/opencode.
• Creed: "personal context file for every agent" product, portable persona/context across agent tools.
• CI pipeline for larger Claude Code projects: community reminder that CI becomes load-bearing once Claude Code projects grow past trivial size.

Agentic SaaS

• Orca: desktop/mobile ADE for running Codex/Claude Code/OpenCode/Pi in parallel worktrees with account switching and usage tracking, direct competitor to herdr/tmux multiplexing setups.
• OmniRoute: free MIT AI gateway aggregating 268+ providers/500+ models with quota-aware fallback and token-compression claims (15-95%), useful if you want to stack free-tier quotas across providers.
• Lunen.ai: team-facing agent-building platform, generic entrant in the no-code agent space.

CLI/TUI & infra

• FastMCP: Pythonic MCP server/client framework taking servers from prototype to production, standard reference if you build MCP tools.
• SnapState: persistent state layer for AI agent workflows.
• cognee: open-source self-hosted knowledge-graph memory platform giving agents persistent cross-session recall, alternative to Creed-style context files.

Vibe-coding & learning

• agency-agents: native desktop app installing a roster of "personality" agents (frontend, Reddit community, whimsy, reality-check) into Claude Code/Cursor/Codex/Gemini.
• ai-job-search: Claude Code-based job-application framework, evaluates postings, tailors CVs/cover letters, runs locally.
• ai-engineering-from-scratch: 503-lesson curriculum (Python/TS/Rust/Julia) targeting the AI-tool-adoption skills gap in students.