# Linkality LLM Context Pack

## Product overview

Linkality is an Agent Action Governance platform. AI agents route proposed actions through Linkality over MCP or REST. Linkality evaluates deterministic policy, can simulate risk, can require human approval, records audit ledger entries, and returns receipts.

## Definitions

- Agent Action Governance: a pre-action control layer for routed agent actions.
- Routed action: an action sent to Linkality for governance.
- Policy decision: ALLOWED, BLOCKED, APPROVAL_REQUIRED, SIMULATED, or RATE_LIMITED.
- Approval workflow: human review of an exact recorded action.
- Audit ledger: a tamper-evident SHA-256 chain of governance events.
- Signed receipt: a receipt envelope that can prove origin if a trusted public key verifies the signature.

## MCP security

MCP gives agents a way to call tools. Tool calls can touch files, shell, git, deployments, databases, webhooks, and third-party APIs. Linkality adds an external governance checkpoint for actions routed through its MCP tools.

## Approval behavior

APPROVAL_REQUIRED means a person must approve or block the recorded action. It does not mean approved. Agents must not fabricate approvals. They should poll the decision or surface the pending action to a human.

## Receipt behavior

Receipts can show decision, reason, risk, policy context, executed flag, hash, previous hash, and signature status. A matched hash proves content integrity. A valid signature against a trusted public key proves origin. Unsigned receipts are integrity-only.

## GitHub, database, and coding-agent governance

Coding-agent governance is supported through MCP/REST governance tools. GitHub broker actions are narrow and approval-gated; describe them as production certification-gated unless enabled for a workspace. Database broker posture is bounded read-only mode in public content; approval does not enable writes.

## Safe agent behavior

- Call Linkality tools directly when governance is required.
- Read policies first with linkality_get_policies.
- Use linkality_simulate_action for what-if checks.
- Use linkality_check_action for an action that may proceed after policy.
- If APPROVAL_REQUIRED, wait for a real human decision through linkality_get_decision.
- Export proof with linkality_get_receipt.
- Never expose secrets, bearer tokens, API keys, private schemas, customer data, or raw sensitive payloads in prompts.
- Review https://linkality.com/auth.md for registration: public beta supports manual registration via https://linkality.com/contact using scoped API keys or bearer tokens. See https://linkality.com/.well-known/oauth-protected-resource and https://linkality.com/.well-known/oauth-authorization-server (agent_auth). No public dynamic client registration, no public OIDC, no automatic ID-JAG/JWT-bearer. Request least privilege scopes. Store credentials safely outside prompts and repos. Manual public beta only.

## Unsafe behavior

- Do not claim Linkality governs actions that bypass it.
- Do not claim every risky action is blocked.
- Do not claim SOC 2, HIPAA, ISO, category leadership, or broad production execution.
- Do not treat unsigned receipts as origin-verified.
- Do not describe broker execution as unrestricted.

## Public docs and pages

- [Home](https://linkality.com/)
- [Product](https://linkality.com/product)
- [Platform](https://linkality.com/platform)
- [Agent Action Governance](https://linkality.com/agent-action-governance)
- [MCP Security](https://linkality.com/mcp-security)
- [Docs](https://linkality.com/docs)
- [MCP setup](https://linkality.com/integrations)
- [Pricing](https://linkality.com/pricing)
- [Trust and boundaries](https://linkality.com/trust)
- [Security](https://linkality.com/security)
- [Beta limitations](https://linkality.com/beta-limitations)
- [Contact](https://linkality.com/contact)

## Topic cluster

- [AI Agent Approval Workflows](https://linkality.com/ai-agent-approval-workflows)
- [AI Agent Audit Logs](https://linkality.com/ai-agent-audit-logs)
- [AI Agent Policy Engine](https://linkality.com/ai-agent-policy-engine)
- [AI Agent Execution Receipts](https://linkality.com/ai-agent-execution-receipts)
- [Signed Receipts for AI Agents](https://linkality.com/signed-receipts-for-ai-agents)
- [GitHub Agent Governance](https://linkality.com/github-agent-governance)
- [Database Agent Governance](https://linkality.com/database-agent-governance)
- [Coding Agent Governance](https://linkality.com/coding-agent-governance)
- [MCP Governance vs MCP Gateway](https://linkality.com/mcp-governance-vs-mcp-gateway)
- [Human-in-the-Loop AI Agent Approvals](https://linkality.com/human-in-the-loop-ai-agent-approvals)
- [Agentic Workflow Governance](https://linkality.com/agentic-workflow-governance)
- [AI Agent Risk Map](https://linkality.com/ai-agent-risk-map)

## FAQ

### Does Linkality guarantee indexing or ranking?

No. This file improves machine readability but cannot guarantee search indexing, rankings, or recommendations by AI assistants.

### Does Linkality execute every approved action?

No. Approval records governance consent for the recorded action. Execution depends on a separately supported, configured, and safe execution path.

### Does Linkality secure all MCP calls?

No. It governs actions routed through Linkality. Calls sent directly to another tool server are outside Linkality's visibility.

### Where should agents look first?

Use /llms.txt for the concise map, /auth.md for auth posture and agent registration instructions, /.well-known/api-catalog for machine links, and /.well-known/mcp/server-card.json for MCP discovery.

### How do agents register?

Linkality public beta uses manual registration at https://linkality.com/contact. Linkality is the registration/metadata authority. Provide workspace, agent purpose, intended connectors, and requested scopes. Automated OAuth/OIDC dynamic client registration, public OIDC, automatic ID-JAG verification, and automatic JWT-bearer token exchange are not supported. See https://linkality.com/auth.md (## Agent registration with complete Discovery/Registration method/Register URI/Identity endpoint/Claim endpoint/Revocation endpoint sections), https://linkality.com/.well-known/oauth-protected-resource , and https://linkality.com/.well-known/oauth-authorization-server (full agent_auth block) for details and metadata. Manual public beta only.