Automaze

 We just shipped CCPM v2 – and it's no longer a Claude Code thing. When we built CCPM a year ago, it was a set of slash commands for Claude Code: /pm:prd-new, /pm:issue-start, the works. It got traction, got starred, got forked. Then people started asking: "Does this work with Cursor?", "What about Codex?", "Can I use it with Amp?" The answer was always: "No." It was a Claude Code-only install via a bash script, and frankly, the install was a mess. So we rebuilt it from scratch as a proper Agent Skill – an open standard that any AI coding harness can load. Same workflow, zero lock-in. What CCPM does: Turn ideas into shipped code with full traceability at every step: PRD → Technical Epic → GitHub Issues → parallel agents executing in separate work streams simultaneously. The discipline: every line of code traces back to a spec. No vibe coding. What changed in v2: ✓ Works with Factory, Claude Code, Amp, OpenCode, Codex, Cursor, and anything else that supports Agent Skills ✓ Deterministic operations (standup, status, search) run as bash scripts – fast, no LLM token cost ✓ Agentic operations (PRD writing, parallel analysis) use the LLM where it actually adds value ✓ No install scripts. Git clone. That's it. ✓ Tested: 100% eval accuracy vs 27.7% baseline without the skill The v1 slash commands are preserved on the v1 branch for anyone who wants them. If you're building with AI agents and are tired of losing context between sessions, this is for you. ⭐ github.com/automazeio/ccpm

We're open-sourcing Proof — a capture SDK and CLI that records visual evidence of test execution. The idea is simple: a 10-second recording of your tests passing is worth more than a 200-line diff. Proof runs your tests and produces shareable artifacts: ✓ Terminal recordings as self-contained HTML players (play/pause, speed control, ANSI colors, zero dependencies, works offline) ✓ Browser recordings via Playwright with cursor highlighting ✓ Structured JSON manifests and markdown reports It works with any test runner — pytest, go test, cargo test, bun test, anything that writes to stdout. Use it from TypeScript, the CLI, or pipe JSON for multi-capture CI runs. We also built an Agent Skill that teaches AI coding agents how to use Proof. Without it, agents default to `tee`, `script`, or raw Playwright runs. With it, they use `proof capture` correctly every time. Benchmark results (Claude Sonnet 4.6, 3 scenarios, 14 assertions): • With skill: 100% pass rate (14/14) • Without skill: 28% (4/14) That's a +72 point improvement from a single skill file. Apache 2.0 licensed. 𝚗𝚙𝚖 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 -𝚐 @𝚊𝚞𝚝𝚘𝚖𝚊𝚣𝚎/𝚙𝚛𝚘𝚘𝚏 https://lnkd.in/eAG6Aa3Q

Would you hire developers in 2026, when AI can build apps in minutes? ...Should you?