Pular para o conteúdo
TakoAPI
Voltar às skills

shieldcortex

por jarvis-drakon

Web & Frontend Development
657Downloads informados pelo clawskills.sh
1Visualizações
0 curtidaEntre para curtir esta skill
—Ainda sem avaliações
Sua avaliação

Resumo

Security framework for AI agents.

Descrição

O que esta skill faz

Persistent memory system for AI agents backed by SQLite with semantic search, knowledge graphs, decay-based forgetting, and contradiction detection. Every memory write passes through a 6-layer defense pipeline that blocks prompt injection, credential leaks, and poisoning attacks. Iron Dome adds behavioral protection with configurable action gates, security profiles, and a full forensic audit trail.

Combines persistent semantic memory and security defense in one package, eliminating the need to wire separate memory and threat-detection systems together.

Quando usá-la

  • Recalling architecture decisions from a previous coding session
  • Detecting prompt injection patterns in agent instruction files
  • Blocking accidental API key writes to agent memory
  • Persisting user preferences across multiple project workspaces
  • Auditing all memory reads and writes for a security review

Exemplo de fluxo de trabalho

ENTRADA

User asks: Remember that the payment API requires OAuth2 bearer tokens, not API keys

AGENTE
  1. 1

    Receive memory write request with content and project scope

  2. 2

    Run content through the 6-layer defence pipeline to check for injection patterns and credential leaks

  3. 3

    Store approved memory in SQLite with importance level and category set to 'architecture'

  4. 4

    Extract knowledge graph entities and relationships from the memory content

  5. 5

    Check for contradictions with existing memories and confirm storage

SAÍDA

Memory stored under category 'architecture' with entities ['payment API', 'OAuth2'] and a relationship triple added to the knowledge graph

Instalação

clawhub install jarvis-drakon/shieldcortex

Execute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.

Links