The CoSAI Risk Map is a framework for identifying, analyzing, and mitigating security risks in Artificial Intelligence systems. As traditional software security practices are not always sufficient for AI, this project provides a shared understanding and a common language for addressing the unique security challenges of the AI development lifecycle.
Skill Claude CodeCodex
Check whether a CoSAI Risk Map entry is pitched at the right altitude (granularity). For a control: objective-not-implementation, not-a-restated-risk, posture-not-mandate, solved-problem, no-duplication. For a risk: the merge-vs-distinct two-test, threat-not-control-gap, and real-not-hypothetical. For a component: the…
Skill Claude CodeCodex
Audit framework mappings — an existing entity's mappings, the whole corpus, or a candidate value proposed for an entity not yet in the corpus — against the framework mappings style guide. Use when reviewing, auditing, or pre-PR authoring mapping changes for risks.yaml, controls.yaml, or personas.yaml.
Skill Claude CodeCodex
Audit persona identification questions in personas.yaml against the identification questions style guide. Use when reviewing or proposing changes to identificationQuestions.
Skill Claude CodeCodex
Ground CoSAI Risk Map terminology in established security terms of art. Use when authoring or critiquing a Control, Risk, Component, or Persona title or description to check a proposed term against the canonical (NIST-first) vocabulary, replace an invented term with its established equivalent, generalize a…
Skill Claude CodeCodex
Select the structured references and framework mappings for a CoSAI Risk Map control or risk — for a control: which components it applies to, which risks it addresses, and which mappings (MITRE ATLAS mitigations, NIST AI RMF subcategories, OWASP LLM, EU AI Act articles) fit; for a risk: which components it impacts…
Skill Claude CodeCodex
Draft a structured maintainer review comment for a CoSAI Risk Map content-proposal issue (a proposed new or updated risk, control, component, or persona), written to a local markdown file for the maintainer to edit and post. Use when reviewing a GitHub content-proposal issue and you want a ready-to-edit review draft.…
Skill Claude CodeCodex
Explain a specific CoSAI Risk Map entry — a risk, control, component, or persona (by id or name) — in plain language for someone new to the framework, including its classical roots and how it connects to other entries. Read-only. Use when someone asks "what is ?", "explain this risk/control", or wants to understand an…
Skill Claude CodeCodex
Answer "what CoSAI controls are like ?" — map an established security idea (e.g. policy enforcement point / PEP, least privilege, reference monitor, defense in depth, zero trust, attestation, provenance) to the CoSAI Risk Map controls that embody or extend it, and explain the classical→AI bridge. Read-only exploration…
Skill Claude CodeCodex
Answer "what risks and controls attach to in an agentic/AI system?" (e.g. AWS Nitro Enclaves, Pinecone as my vector store, LangChain orchestration, or a component id directly). Maps the product/technology to the CoSAI component(s) it implements or protects, then surfaces that component's exposure — the controls that…
Skill Claude CodeCodex
Show what the CoSAI Risk Map maps to a given external-framework entry — e.g. "everything under EU AI Act Article 14", "which risks map to MITRE ATLAS AML.T0051", "our NIST AI RMF GOVERN-6.2 coverage", "STRIDE Tampering", "ISO 22989 AI Producer". A read-only reverse index over the mappings across risks, controls, and…
Skill Claude CodeCodex
Help someone figure out which CoSAI Risk Map persona(s) they are — by walking the framework's identificationQuestions — and then show what that role should care about: the risks that impact them and the controls they are responsible for implementing. Read-only and guided. Use when someone asks "which persona am I?"…
Skill Claude CodeCodex
Answer "what risks am I exposed to if I ?" — map a person's activity or role (e.g. "I develop agents", "I use an AI coding assistant", "I host third-party models") to the CoSAI Risk Map risks that affect them, plus the controls that address those risks. Read-only, for security architects, developers, and technology…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: