Trusted — Risk Score 0/100
Last scan:2 days ago Rescan
0 /100
llm-knowledge-bases
Inspired by a public workflow shared by Andrej Karpathy (@karpathy). From raw research to a living Markdown knowledge base that compounds with every question.
LLM Knowledge Bases is a lightweight, documentation-only skill that defines workflows and rules; all actual Vault I/O is delegated to a separate plugin, with no declared or inferred sensitive capabilities beyond documentation metadata.
Skill Namellm-knowledge-bases
Duration29.2s
Enginepi
Safe to install
Approve for use. No security concerns identified.
ResourceDeclaredInferredStatusEvidence
Filesystem NONE NONE SKILL.md explicitly forbids direct file writes; all Vault I/O is mediated by plu…
Network NONE NONE No network calls; only references https://x.com/karpathy in attribution text (RE…
Shell NONE NONE Shell scripts only scaffold directories (mkdir/printf) and invoke clawhub CLI wi…
Environment NONE NONE No environment variable access in any script or implementation
Skill Invoke NONE NONE Skill defines workflows but does not invoke other skills
Clipboard NONE NONE No clipboard access found
Browser NONE NONE No browser access found
Database NONE NONE No database access found
1 findings
🔗
Medium External URL 外部 URL
https://x.com/karpathy
README.md:8

File Tree

10 files · 22.5 KB · 773 lines
Markdown 6f · 690L Shell 2f · 62L JSON 1f · 17L YAML 1f · 4L
├─ 📝 CHANGELOG.md Markdown 47L · 2.3 KB
├─ 📋 clawhub.json JSON 17L · 439 B
├─ 🔧 init_llm_kb_repo.sh Shell 51L · 1.2 KB
├─ 📝 maintenance-playbook.md Markdown 91L · 2.6 KB
├─ 📋 openai.yaml YAML 4L · 272 B
├─ 🔧 publish.sh Shell 11L · 465 B
├─ 📝 README.md Markdown 158L · 4.7 KB
├─ 📝 RELEASE.md Markdown 74L · 2.7 KB
├─ 📝 repo-layout.md Markdown 110L · 2.5 KB
└─ 📝 SKILL.md Markdown 210L · 5.4 KB

Security Positives

✓ All capabilities are explicitly NONE; skill is a documentation-only specification
✓ SKILL.md clearly states: do not use generic file tools for Vault operations — only plugin tools
✓ Shell scripts use only safe primitives (mkdir, printf) with hardcoded paths and no pipe from user input
✓ No external dependencies (no requirements.txt, package.json, Cargo.toml, or Pipfile)
✓ Safety boundaries are well-defined: never modify raw/, never write wiki/ directly, never fabricate IDs
✓ Failure handling is explicit and prevents bypass: do not work around the error by writing directly to the Vault
✓ No base64, no eval, no curl|bash, no environment iteration, no credential access
✓ No hidden HTML comments, no steganography, no obfuscation
✓ No sensitive paths (~/.ssh, ~/.aws, .env) accessed anywhere