{"slug": "show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate", "title": "Show HN: Curated Claude Code – a small agent harness with an intake gate", "summary": "A developer released Curated Claude Code, a small agent harness for Claude Code that prioritizes safety and self-evolution over bloated skill collections. The system features an intake gate (/vet) that blocks unverified tools, a four-layer learning loop (corrections → memory → skills → gated intake), and no auto-running hooks to prevent unauthorized mutations. The repository includes a curated set of skills and agents designed to compound improvements over time without context bloat.", "body_md": "**A small, vetted, self-evolving harness for Claude Code — curated, not dumped.**\n\n## 🇷🇺 Кратко на русском\n\nНебольшой курированный харнесс для Claude Code — не свалка из сотен блоков, а горстка скиллов, которые заслужили место, плюс каркас безопасности и (главное) **цикл самоулучшения**: правки пользователя становятся памятью, находки — скиллами, вход нового инструмента идёт через гейт `/vet`\n\n(безопасность — блокирующий шаг). Никаких авто-хуков, которые мутируют что-то без спроса. Подробности — в [ docs/EVOLUTION.md](/Sanexxxx777/curated-claude-code/blob/main/docs/EVOLUTION.md). Полный текст ниже на английском.\n\nMost \"everything for Claude Code\" collections optimize for **count** — hundreds of agents, skills, rules, and event hooks that auto-run on your tool calls. It looks impressive and fails quietly: bloated context, overlapping triggers fighting to activate, and hooks that mutate things without asking.\n\nThis repo optimizes for the opposite. A handful of skills that earn their place, a spine of safety guards, and — the actual product — **a system that improves itself over time, safely.**\n\nA harness that grows is worth more than a harness that's big.The thing that compounds isn't the block count — it's the intake gate, the hygiene loop, and the learning machinery. →is the heart of this repo; start there.`docs/EVOLUTION.md`\n\n- ✅ A curated set of executable skills + always-on principles you can drop into\n`~/.claude/`\n\n. - ✅ A\n**method**(`/vet`\n\n) for deciding what to adopt, and a registry so you never evaluate the same tool twice. - ✅ A\n**self-evolution loop**— corrections become memory, findings become skills, the stack stays clean. - ✅\n**Safety by design**— no surprise autonomous actions. - ❌ Not a 500-block mega-bundle. Not a framework you fork. No auto-running hooks.\n\nSingle-skill self-learning harvesters are trending, and the good ones are genuinely good. This repo is the **system that pattern lives in**: four learning layers (corrections → memory → skills → gated intake) instead of one harvest step, a promotion gate so unverified guesses never become authoritative skills, an intake gate (`/vet`\n\n) so the stack doesn't bloat as it learns, and a hygiene loop (`system-health`\n\n) so it doesn't rot. If you only want the harvest step, [ Kulaxyz/self-learning-skills](https://github.com/Kulaxyz/self-learning-skills) does it well — we credit and cherry-pick it in\n\n[. If you want learning as an](/Sanexxxx777/curated-claude-code/blob/main/docs/EVOLUTION.md)\n\n`docs/EVOLUTION.md`\n\n*operating discipline*, start here.\n\n**Vet before adopt.** Nothing enters on hype. Every incoming tool runs through one pipeline — relevance →**security as a blocking gate**→ duplication → real ROI → verdict. The security phase reads a skill's body as a*behavioral*instruction and checks the agentic risks (hidden instructions, prompt injection, trigger abuse, excessive agency) a code-only scan misses.**It evolves.** A user correction is the strongest learning signal; a finding is captured at discovery; a 5+-iteration procedure becomes a skill — but only past a**promotion gate**(verified check + named failure pattern + a ruled-out dead-end), so confident guesses never get enshrined as truth. The system compounds instead of going stale. →`docs/EVOLUTION.md`\n\n**Cherry-pick over giant.** A good idea inside a huge framework doesn't justify installing the framework. Extract the technique, leave the bloat.**No auto-hooks.** Auto-mutation without confirmation is exactly the failure mode to avoid. →`rules/safety-guards.md`\n\nFull rationale: [ docs/METHODOLOGY.md](/Sanexxxx777/curated-claude-code/blob/main/docs/METHODOLOGY.md).\n\n| Skill | What it does |\n|---|---|\n`vet` |\n\n`workflow-upgrade`\n\n`goal`\n\n`konsilium`\n\n`ship-secure`\n\n`pre-push`\n\n`system-health`\n\n`teach`\n\n`longread`\n\nTwo by design, not sixty — each reflects the verify-before-trust stance, not filler.\n\n| Agent | What it does |\n|---|---|\n`adversarial-verifier` |\n\n`tool-scout`\n\n`/vet`\n\n— reads a tool's repo/docs and returns a structured security report; never installs or runs.| Rule | What it covers |\n|---|---|\n`code-principles.md` |\n\n`safety-guards.md`\n\n`root-cause-discipline.md`\n\n`new-project.md`\n\n`detective-mindset.md`\n\n—`docs/EVOLUTION.md`\n\n**how the harness improves itself**(start here).— why curated, not dumped.`docs/METHODOLOGY.md`\n\n— a file-based, self-learning memory pattern.`docs/MEMORY_SYSTEM.md`\n\n— task → tool routing, proactively.`docs/TOOL_ROUTING.md`\n\n— backup → surgical → verify → record.`docs/CHANGE_DISCIPLINE.md`\n\n— a registry template and a worked`examples/`\n\n`/vet`\n\nverdict.\n\n| I want to… | Use |\n|---|---|\n| Decide whether to adopt a new tool/skill/MCP | `vet` |\n| Run a deliberate pass to improve my whole setup | `workflow-upgrade` |\n| Hand a goal to the agent and get back a finished, verified result | `goal` |\n| Pressure-test a high-stakes decision (go/no-go) | `konsilium` |\n| Make sure a web app is safe before launch/handoff | `ship-secure` |\n| Make sure nothing junk leaves on push/deploy | `pre-push` |\n| Keep my Claude Code config from rotting | `system-health` |\n| Actually understand a session/article/concept | `teach` |\n| Write an analytical longform piece | `longread` |\n\nInstall manually — copy only what you want. (No `curl | bash`\n\n: this harness practices what it preaches.)\n\n```\ngit clone https://github.com/Sanexxxx777/curated-claude-code\ncd curated-claude-code\n\n# install the skills you want\ncp -r skills/vet ~/.claude/skills/\ncp -r skills/ship-secure ~/.claude/skills/\n# …repeat for the rest\n\n# agents → ~/.claude/agents/ ; rules → append into your CLAUDE.md or ~/.claude/rules/\ncp agents/*.md ~/.claude/agents/\n```\n\nRestart Claude Code, then invoke a skill by its name (e.g. `/vet`\n\n).\n\nThis harness was built the way it tells you to build — by cherry-picking techniques and crediting them:\n\n`konsilium`\n\nadapts Andrej Karpathy's(independent opinions → anonymized peer review → chairman synthesis).`llm-council`\n\n`ship-secure`\n\nadapts Prajwal Tomar's \"Vibe Coders Are Getting Sued\" launch playbook.`code-principles`\n\nis derived from Karpathy's widely-shared CLAUDE.md principles.`goal`\n\n's design phase cherry-picks loop-design ideas from.`ksimback/looper`\n\n`teach`\n\nadapts Anthropic's guided-learning prompt.- The skill promotion gate in\n`EVOLUTION.md`\n\ncherry-picks the verification bar from.`Kulaxyz/self-learning-skills`\n\nI build and harden Claude Code workflows, agentic setups, and web apps — and I do managed hosting and site builds. If these patterns are useful to your team, or you want help applying them, reach out:\n\n**GitHub:** open an issue, or[@Sanexxxx777](https://github.com/Sanexxxx777)**Email:**[sanexxx777@gmail.com](mailto:sanexxx777@gmail.com)\n\nPRs welcome — see [ CONTRIBUTING.md](/Sanexxxx777/curated-claude-code/blob/main/CONTRIBUTING.md). The bar: it must pass its own\n\n`/vet`\n\n, ship no auto-hooks, contain no secrets, and stay surgical. Curation is the point; volume isn't.If this saved you time, a ⭐ helps others find it.\n\n[MIT](/Sanexxxx777/curated-claude-code/blob/main/LICENSE) © Aleksandr Shulgin ([@Aleksandr_NFA](https://t.me/Aleksandr_NFA)) · [@Sanexxxx777](https://github.com/Sanexxxx777)", "url": "https://wpnews.pro/news/show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate", "canonical_source": "https://github.com/Sanexxxx777/curated-claude-code", "published_at": "2026-07-25 11:14:36+00:00", "updated_at": "2026-07-25 11:22:16.201878+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-safety", "developer-tools"], "entities": ["Claude Code", "Curated Claude Code", "Sanexxxx777", "Kulaxyz/self-learning-skills"], "alternates": {"html": "https://wpnews.pro/news/show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate", "markdown": "https://wpnews.pro/news/show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate.md", "text": "https://wpnews.pro/news/show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate.txt", "jsonld": "https://wpnews.pro/news/show-hn-curated-claude-code-a-small-agent-harness-with-an-intake-gate.jsonld"}}