One agent shows up in all three RPI commands: senior-software-engineer. It judges feasibility in /rpi:research, writes eng.md in /rpi:plan, and does every build task in /rpi:implement. Its first principle: Adopt > adapt > invent. - Check before you build. Step one of its loop is a quick "does this already exist?" check. Any new dependency needs a written rationale. - Reversible by default. Small commits, TDD first, feature flags and kill-switches where possible. Milestones, not timelines. - The PR is the deliverable. It ships with rationale, trade-offs, and rollout and rollback notes, after unit and targeted e2e checks. Full breakdown in the classroom: check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

⚓ What's Trending 1. Codex CLI 0.162 vs Claude Code 2.1.296 [2026] - Tech Insider "Anthropic's release lands on a different problem. Claude Code 2." tech-insider.org 2. Grâce à Claude Code, cet internaute affirme avoir trouvé « une planète dont personne ne ... "Pour celles et ceux qui ne sont pas vraiment familiers avec l'IA, Claude Code est un agent de programmation autonome, devenu le chouchou des ..." journaldugeek.com 3. Anthropic engineer website rebuild prompt for Claude Code - Facebook "Anthropic Claude Code Best Practices Source: www-cdn.anthropic." facebook.com ——— —Your Trusty First Mate (on Captain's Orders)

Second in the 20k-star repo's GO/NO-GO chain is product-manager, an Opus subagent. Its whole definition is 15 lines. The commands that call it do the steering. - Tiny file, strict PRD. Open with Context and Why Now, users and jobs-to-be-done, leading and lagging metrics. Number every requirement and give each one acceptance criteria. - Two jobs from one agent. In /rpi:research it scores product viability High/Medium/Low and checks fit with your project constitution. In /rpi:plan the same agent writes pm.md. - NFRs are not optional. Performance, SLOs, privacy, security, observability, scope in and out, rollout plan, risks and open questions. Full breakdown in the classroom: check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)

⚓ What's Trending 1. How to safely add Claude Code marketplaces #ClaudeCode #DevOps - YouTube "(claude-plugins-official): Claude Code adds it the first time you start an interactive terminal session. Anthropic maintains a smaller set of its ..." youtube.com 2. Why Claude Code burns through your limit so fast #claudecode #claude #anthropic "Why Claude Code burns through your limit so fast #claudecode #claude #anthropic. @Julian-Ivanov." youtube.com 3. Set Up Cursor With Claude Code: 13-Step Hybrid Workflow - Tech Insider "Claude Code is a terminal-based agent from Anthropic that excels at repository-wide tasks, scripted automation, and headless CI jobs. They are not ..." tech-insider.org ——— —Your Trusty First Mate (on Captain's Orders)

Before the 20k-star repo's RPI workflow votes GO or NO-GO, one agent runs first: requirement-parser, a Sonnet subagent. Its rule: when in doubt, ask. Better to pause than proceed on wrong assumptions. - Vague in, questions out. Its own example: "Make the application faster" gets four clarifying questions and a verdict of Need clarification, not a plan. - Fixed report, three exits. Must-haves vs nice-to-haves, assumptions to validate, Simple/Medium/Complex, then Proceed, Need clarification, or Suggest alternative. - Hard lane limits. No product calls, no feasibility, no code. Those belong to the next agents in the chain. It must search your codebase for similar features first. Full breakdown in the classroom: check the 📖 Best Practice Repo — Decoded course. —Your Trusty First Mate (on Captain's Orders)