Restrict the sources.
Choose permitted websites and providers. Strict mode blocks reads outside your policy.
Your employees have access rules.
Your agents should too.
Apply access and licence rules. Check incoming content before your agents read it through Common Measure.
Check the content.
Run checks for unwanted instructions and personal data before content reaches the model. Developers can add your organisation’s own checks.
Apply the team’s rules.
Distribute source rules through Hub. Keep each agent’s decisions and reasons on record.
Control copyright
and contract risk.
A subscription or API key alone does not establish permission to reuse the material it returns.
How source terms are checkedRequire the agreed licence.
In strict mode, refuse reads when the licence required by your policy is missing or does not match.
Keep the basis for use on record.
Keep the source’s terms alongside your organisation’s assessment of the agreement, public licence or exception covering the intended use.
A licence does not establish accuracy or protect against prompt injection and poisoning.
A blocked read,
with the reason.
A recorded test of instructions hidden in internal documents to redirect the agent.
The screen checks known patterns; passing it does not establish that content is safe.
Inspect the full recordDocumentStrict mode
- connecting a data sourceAllowed
No rules matched. Passed to the agent.
- community answer threadBlocked
3 instruction patterns matched. Kept out of the agent’s context.
Before the read
Reads sent through Common Measure can be refused before content enters the agent.
After the read
Tools that report afterwards provide a record, without blocking the read.
Other tools
Built-in tools are covered only where they are routed through Common Measure, as Claude Code’s WebFetch and WebSearch are with the plugin. Other built-in tools and hidden context are not. Keep your device protection and restrictions on what agents can do.