Skip to content

v1.4.0 — VKirill adaptations + TGApp skills + GAN harness

Choose a tag to compare

View all tags
@RaNDoM6913 RaNDoM6913 released this 14 May 01:21
v1.4.0
07047ae

🎯 Highlights

Substantial expansion across both Claude Code and Codex CLI. Four themes:

🧠 VKirill/codex-starter-kit adaptations (3 specialist roles + Codex DSL)

  • minimal-change-engineer — surgical implementation. Value measured in lines NOT written. Refuses scope creep, prefers three similar lines over a premature abstraction.
  • reality-checker — defaults to NEEDS WORK, refuses fantasy A+ ratings. Demands screenshots, logs, test outputs.
  • codebase-onboarding-engineer — produces a concise 30-60 min brief for unfamiliar codebases.
  • intake-classifier.py — UserPromptSubmit hook with deterministic scoring (0-15) on RU+EN keywords + optional gpt-5.5-nano LLM fallback when confidence < 0.78.
  • packages/codex/rules/default.rules — Starlark-like DSL for Codex CLI approval policy (~380 lines). Forbids destructive ops, allows safe dev commands, prompts on risky ones.

📱 TGApp / general production skills

All dual-format (Claude Code + Codex CLI):

  • telegram-bot-builder — Telegraf, grammY, aiogram patterns, monetization, webhook vs polling
  • nextjs-supabase-auth — Next.js 14+ App Router + @supabase/ssr, RLS, common production issues
  • drizzle-orm-expert — TypeScript-first ORM, SQL-like + relational APIs, edge-runtime patterns
  • ru-text — Russian typography: «», em dash, NBSP, ellipsis, decimal comma, № sign, ₽, 14 stop-words to avoid
  • postgresql-optimization — 7-phase workflow with EXPLAIN node interpretation, index selection guide
  • redis-patterns — Cache-aside, pub/sub, BLPOP queue, distributed locks with Lua, eviction policies
  • behavioral-nudge-engine — retention psychology, Fogg Behavior Model, habit loops, anti-pattern table

🛡️ Competitor patterns from everything-claude-code (181k stars)

  • GateGuard hooks — gateguard-pre-edit.sh + gateguard-record-facts.sh. Require Grep/Read before Edit/Write/Bash. Read-only Bash exempt. Strict mode (CLAUDE_GATEGUARD_STRICT=1) blocks.
  • silent-failure-hunter expanded — 109 → 250 lines. 6-category taxonomy (empty handlers / promise suppression / fallback masking / log-and-forget / generic catch-all / linter suppression) + per-language BEFORE/AFTER fix examples (TS, Python, Go, Bash).

🎯 GAN harness package (new, optional)

  • packages/gan/ — three-agent adversarial verification loop (gan-planner → gan-generator → gan-evaluator)
  • Two rubric files (ui-quality.md with 17 binary criteria, functionality.md with 15)
  • Anti-AI-slop auto-fails: console.log, "something went wrong", placeholder text, no empty state
  • Requires Playwright (~150 MB) — opt-in install

🐹 Go references +5 (24 → 29)

  • di-frameworks.md — uber-fx, uber-dig, google-wire comparison
  • graphql-patterns.md — gqlgen schema-first workflow
  • module-management.md — go.mod/go.sum/workspaces
  • stay-updated.md — Go release cadence + recent additions
  • standard-stdlib-now.md — what stdlib now replaces

📊 Counts after v1.4.0

Component Before (1.3.11) After (1.4.0) Δ
Core agents 27 31 +4
Core skills 5 11 +6
Core hooks 25 28 +3
Codex skills 72 82 +10
Codex rules 0 1 file +1
Go references 24 29 +5
Packages 7 8 (+gan) +1
Total agents 49 56 +7

📥 Install

npx claude-code-superkit
# Or upgrade existing install:
npx claude-code-superkit upgrade

🔍 Source attribution

Full diff: v1.3.11...v1.4.0

🤖 This release was authored by Claude Opus 4.7 (1M context) and will be reviewed in a separate Codex CLI (gpt-5.5) session by the maintainer to catch issues the authoring model missed.

Assets 2
Loading