Agent-native market data

Read-only market context before trading decisions

Read-only context gives agents current market evidence without granting order authority. Cornerstones should inform analysis; a separate trading runtime should own risk and execution.

Compare plans Docs

Question

Why should trading agents fetch read-only context before they touch execution or risk controls?

When this problem happens

Agents can overreach if market data and execution tools live behind the same unchecked interface.
Human reviewers need to reconstruct what context existed before a recommendation.
Market conditions change, so copied examples are not enough for serious workflow review.

Recommended architecture

Keep market reading, risk evaluation, and execution as separate boundaries, with Cornerstones responsible only for context.

Implementation steps

Fetch context through Cornerstones before any strategy or order discussion.
Record freshness, market surface, and plan boundary in the analysis artifact.
Run risk, sizing, approval, and execution only in the customer trading runtime.
Review context drift before reusing an old agent recommendation.

Example workflow

cornerstones-client verify
# persist context evidence before any execution workflow starts

Comparison table

Option	Best for	Tradeoff
Read-only context first	Safe analysis and audit trails	Requires separate execution workflow
Combined read/write tool	Highly controlled internal automation	Higher blast radius for agent mistakes
Static screenshots	Discussion snapshots	Not repeatable or machine-verifiable

FAQ

Does read-only context remove risk?

No. It reduces tool blast radius, but trading decisions still need risk controls and human or system governance.

What should be saved?

Save the market surface, timestamp, plan boundary, and key facts used by the agent.

Can context be premium?

Yes. Premium context is fine when the account plan permits it and execution remains separate.