agentsclimarketplace

Auto capability acquisition

Skill alidevai/chiron_agent/.claude/skills/auto-capability-acquisition

Autonomous, self-improving AI agent platform for Claude Code — safe capability & skill acquisition, MCP/tool discovery, sandboxed policy gates, and token-efficient learning.

Install
npx -y skills add alidevai/chiron_agent --skill auto-capability-acquisition

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Guvenilir kaynaklardan (resmi org veya GitHub'da yuksek yildizli, bakimli repo) gelen dusuk/orta riskli yetenekleri INSAN ONAYI BEKLEMEDEN otomatik edinir ve kurar; icerigi Sonnet ile okuyarak guvenlik denetiminden gecirir. Eksik yetenek tespit edildiginde ve aday guvenilir kaynaktan geliyorsa bu otomatik hat kullanilir.

SKILL.md

3.2 KB, as published. Nobody here has run it

Auto Capability Acquisition (guvenilir kaynak = otomatik)

Bu skill, [[secure-capability-acquisition]] akisinin OTOMATIK seridir. Fark: guvenilir kaynaktan gelen adaylar icin insan onayi beklenmez — ama UC bagimsiz guvenlik kapisi yine de zorunludur.

Ne zaman otomatik, ne zaman insan?

Otomatik (insan onayi YOK): aday su kosullari saglıyorsa

  • Kaynak resmi/guvenilir org (anthropics, modelcontextprotocol, microsoft, trailofbits, github, ...) VEYA GitHub yildizi esigin ustunde (varsayilan >=800)
  • Repo bakimli (son push < ~18 ay) ve lisans uygun (MIT/Apache/BSD/...)
  • Risk dusuk veya orta
  • Tehlikeli izin YOK (oauth, broker, production_write, genis dosya sistemi, secret export, email, prod db, main'e push)

Insan onayi (otomatik DEGIL): yukaridakilerden biri saglanmiyorsa — bilinmeyen kaynak, dusuk yildiz, bakimsiz, yuksek risk veya tehlikeli izin. Aday kaybolmaz; onay paketine duser.

Otomatik akis

  1. Aday + metadata topla. capability-manager subagent adayi bulur ve su metadata'yi cikarir (JSON):

    {"id": "...", "source_url": "...", "org": "...", "stars": 0,
 "pushed_days_ago": 0, "license": "MIT", "risk_level": "low", "permissions": []}

  2. Kapi 1 — guven katmani:

    python -m core autoacquire-check <id> --meta-file <metadata.json>

    eligible=false ise otomatik durur -> insan onayi. true ise devam.

  3. Staging + Kapi 2 — deterministik tarama:

    python -m core stage <dizin> --id <id> --risk <low|medium> --source <url>
python -m core eval <id>

    Kritik bulgu = otomatik red. Eval >= 0.90 gerekli.

  4. Kapi 3 — Sonnet icerik incelemesi: auto-security-reviewer subagent'ini cagir (Sonnet modeliyle calisir). Dosyalari okur, JSON verdikt dondurur ve bir dosyaya yazar.

  5. Otomatik kurulum (uc kapi da gecerse):

    python -m core autoacquire-promote <id> --meta-file <metadata.json> \
    --review-file <sonnet-verdict.json>
    • Uc kapi gecti -> otomatik kurulur, audit'e tam koken (kaynak, yildiz, lisans, tarama skoru, Sonnet verdikti) yazilir.
    • Herhangi biri gecmezse -> otomatik kurulmaz, approvals/pending/ altina insan onay paketi duser.

Neden guvenli?

  • Guven katmani tek basina yeterli DEGILDIR; tarama + Sonnet okumasi da sart.
  • Deterministik tarama LLM'e degil regex'e dayanir; dis saldiri onu atlayamaz.
  • Yuksek/kritik risk ve tehlikeli izinler guven katmaninda elenir; sealed immutable-core zaten oauth/broker/high-risk kurulumu insana baglar.
  • Isi bulan (capability-manager) ile denetleyen (auto-security-reviewer) ayridir.

Kaynaklar

  • Motor: core/autoacquire.py, core/lifecycle.py (auto_promote)
  • Ilgili: [[secure-capability-acquisition]], [[mcp-security-review]], [[capability-gap-analysis]], [[minimalist-engineering]]

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.