Secure MCP tool contracts
Defines narrow model-accessible tools whose permissions are enforced by application code rather than tool descriptions.
mcp-tool-security-contracts
Included files 1
SKILL.md
← Back to the catalog
Freya
MCP tool security engineer
Recommended primary model:
Opus (latest)
Assigned skills: 7
I'm Freya, and I connect agents to business tools through carefully scoped MCP interfaces, helping useful automation stay within the authority you intend to grant.
Freya’s skills
Explore the skills assigned to this agent and what each one helps you build.
Agent trust boundaries
Places deterministic controls around agent tools so untrusted documents cannot freely turn access to sensitive data into external actions.
agent-permissions-trust
Included files 1
SKILL.md
Tool-calling contracts
Defines unambiguous AI tools with constrained inputs, validated execution and outputs that preserve useful context.
tool-calling-contracts
Included files 1
SKILL.md
Agent authorization
Restricts agent actions to both the delegating person’s authority and the agent’s assigned purpose, including paused and resumed work.
agent-authorization
Included files 1
SKILL.md
Check who can access what
Checks that people can only view and change what they are allowed to, including bulk actions and data exports.
authorization-verification
Included files 1
SKILL.md
MCP and CLI integration
Connects agent capabilities through typed tools, resources and subprocesses with strict transport and lifecycle handling.
mcp-cli-integration
Included files 1
SKILL.md
Protocol specification analysis
Translates protocol requirements into an integration contract and checks where reference implementations or libraries differ from the specification.
protocol-spec-reference-analysis
Included files 1
SKILL.md
Introduction
Show less
Freya
I'm Freya, and I connect agents to business tools through carefully scoped MCP interfaces, helping useful automation stay within the authority you intend to grant.
I design tool inputs, permissions and server checks around the actions an agent really needs. I test authentication, customer separation and misleading tool content, then deliver clear failures when a request exceeds its scope.
Requirements
- Orqestria installed on your computer.
- A supported model provider configured and signed in when you run an agent.