Forge Fleet 1.0.8 is out.
- sessions and containers named after the env, so /resume shows which is which
- --subagent picks a Claude Code subagent, no template needed
- fixed: "newer image available" on every run under container
Local-first, free: https://t.co/bKfqRuPTpr
Two commands in Claude Code:
claude plugin marketplace add grzegorz-aniol/agent-plugins
claude plugin install forge-fleet@agent-plugins --scope user
Codex uses the same marketplace. Tell me what you would add next.
Forge Fleet now has a skill, so your coding assistant can drive it.
Say "start on issue 355" and you get a sandbox: own worktree, own container, own agent session. It checks for unpushed work before it tears one down.
Local-first, free: https://t.co/bKfqRuPTpr
𝗙𝗼𝗿𝗴𝗲 𝗙𝗹𝗲𝗲𝘁 𝟭.𝟬.𝟳 is out -> https://t.co/FkuqZbJCUQ
- arrow-key picker when you run a command without naming an environment
- Esc cancels a prompt cleanly, no error dump
- ffleet up tells you when a newer agent image is waiting, and offers to pull it @claudeai
A walk-through of a normal working day with Forge Fleet.
One command, and a coding agent is on a @GitHub issue in its own worktree, its own container, its own @tmux session. Then I go do something else.
Feature done, PR reviewed, sandbox gone. @claudeai@OpenAICodexCli
1/ Meet Forge Fleet: one CLI, one sealed environment per coding agent. macOS and Linux, one command to install.
brew tap grzegorz-aniol/tap · brew install ffleet
3/ Spins up a fresh Git worktree, or works right in your existing workspace, your call. Fully loaded with prompt templates for GitHub issues and Linear: point at a ticket, get a ready-seeded environment.
Supergateway has been my secret weapon for months.
I run many #MCP servers in Docker and jump between different agents and transports (stdio, HTTP, SSE). When a server and a client don’t speak the same protocol, Supergateway bridges them 👉 supercorp-ai/supergateway
The Claude Code #SKILL can be easily configured for self-improvement.
Try adding this rule to your skill:
* After each session, do a brief self-reflection and update this skill if any new, important execution rules were learned. You can try it with #Claude Code, #Codex, #Cline
I rely on CLI coding agents a lot. They work well while I’m watching the terminal, but the real friction starts when I switch to other work and wait for them to finish. In my next blog post I show how I solved this in #Claude#Code and #Cline
https://t.co/gscYxHsXSo
@matadcze @KinasRemek Zawód programisty przeobraża się w stronę architekta i projektanta systemów - product engineer. Języki programowania jeszcze nam będą potrzebne do interpretacji wyników, ale docelowo staną się tym czym dzisiaj jest kod maszynowy, już na tym poziomie prawie nikt nie pracuje
I love Claude Code’s /context. It exposes context token usage for your MCP servers, which is exactly the metric we should enforce in CI/CD to avoid gigantic prompts. Example from the GitHub MCP server: https://t.co/GdZ2On3H67
📊
#MCP#Claude#CICD#PromptEngineering
I came across a developer with heavy commits for years, not just one. Respect. It reminded me to ask: where do you recharge, and how much does it boost your performance long term? ☕🎧
Ever noticed ChatGPT seems to know more about you than what you said in a single chat? I was surprised too. It can reuse previous chat summaries to personalize replies. Worth a read if you care about UX and privacy. 🤔
#AI#ChatGPT#Privacy#UX https://t.co/dGS9h1VMD1