Skip to content

v3.0.0 — The Contracts Release

Latest

Choose a tag to compare

@KhazP KhazP released this 18 Jul 21:46

Release Date: July 19, 2026

📦 Release Archive: vibe-coding-prompt-template-v3.0.0.zip (~146 KB)

🌟 Highlights

  • Handoff Context Blocks: research → PRD → tech design docs now end with a machine-readable ## Handoff Context block that the next stage parses — no more re-answering your skill level, budget, and timeline at every step.
  • One Canonical AGENTS.md: templates/AGENTS.md is now the single 10-section template (adds Current State, Roadmap, How I Should Think, What NOT To Do); part4 and all Claude skills reference it — three divergent variants removed.
  • Four-Tool Support: AGENTS.md is the universal contract — Codex reads it natively. Thin adapters ship for Claude Code (CLAUDE.md), Cursor (.cursor/rules/vibe.mdc), and Antigravity (.agent/rules/vibe.md) in templates/tool-adapters/.
  • /vibe- Commands Everywhere:* optional stage commands for Cursor (.cursor/commands/), Antigravity (.agent/workflows/), and Codex (~/.codex/prompts/) — the same guided flow the 6 Claude Code skills provide.
  • part4 Rebuilt: one instantiation workflow (copy templates → fill → single stop), dual IDE/chat entry modes, [REPLACE:]/[CHOOSE:] placeholder rule. The contradictory inline templates are gone.
  • part3 Slimmed 42 KB → 22.5 KB: embedded Terraform/SQL/CI reference code removed; fabricated Cursor settings and stale cursor.sh removed; ## Project Structure added to the Vibe template.
  • No More Fabricated Sources: part1 citations branch on web-search availability — sources with access dates when online, Unverified — model knowledge labels when offline.
  • Hooks That Actually Run: Claude Code hooks moved from the never-loaded .claude/hooks/hooks.json into .claude/settings.json — hardened for Windows paths, .env.example allowed, wider destructive-command coverage.
  • Worked Example: examples/reddit-to-ai/ shows the full paper trail (research → PRD → tech design → AGENTS.md + MEMORY.md) reconstructed from a real shipped project.
  • Self-Checking Repo: scripts/validate.py runs 13 contract checks (template sections, handoff blocks, skill anchors, frontmatter); CI adds lychee link checking.
  • Skills Synced: question banks defer to the root part files (single source of truth); skill outputs/intakes honor Handoff Context; all six declare allowed-tools; vibe-build uses REVIEW-CHECKLIST.md (now with a Security section) and MEMORY.md.
  • ⚠️ Breaking: part4's flow and the AGENTS.md structure changed — re-instantiate your agent files after upgrading. Older docs without Handoff Context blocks still work (stages just ask instead of confirm).

📁 What's Included

  • Core Prompts: All 4 prompt parts (part1–part4, updated).
  • .claude/: 6 skills (vibe-research, vibe-prd, vibe-techdesign, vibe-agents, vibe-build, vibe-workflow), hooks in settings.json, README.
  • templates/: canonical AGENTS.md, MEMORY.md, REVIEW-CHECKLIST.md (with Security section), agent_docs/, tool-adapters/ (CLAUDE.md, Cursor + Antigravity rules, 18 /vibe-* command files).
  • examples/: reddit-to-ai worked example (reconstructed, labeled).
  • docs/: claude-agent-teams (verified-claims rewrite), cursor-cloud-agents, freshness-policy, golden-path-checklist.
  • .github/: workflows (repo-lint with contract checks + link checking, greetings, pr-size-labeler), updated SECURITY.md, issue templates, FUNDING.yml.
  • scripts/: validate.py — run the contract checks locally before committing.
  • Root Files: README.md, CHANGELOG.md, CITATION.cff, LICENSE, .gitignore.