Agent Integration

Give your coding agent (Claude Code, Cursor, Codex, or any tool that reads AGENTS.md/CLAUDE.md) a standing instruction to route risky actions through LetsCompl.ai before executing them. Paste the block below into your repository's AGENTS.md or CLAUDE.md.

Paste-in block

## Compliance Gateway (LetsCompl.ai)

Before executing any of the following, call the LetsCompl.ai gateway first and abort if the verdict is `blocked`:
- Payouts, transfers, refunds, or any write that moves money
- Direct writes to a production database or ledger table
- Any read or write that touches PHI (medical record numbers, diagnoses, patient names) or PII (SSN, full card numbers)

How to call it:
- TS/JS: `import { LetsComplaiClient } from "@letscomplai/sdk"; const result = await client.evaluate(payload);`
- Python: `from letscomplai import LetsComplaiClient; result = client.evaluate(payload)`
- MCP: call the `letscomplai_evaluate` tool with the action payload

If `result.verdict === "blocked"`, do not execute the action — surface `result.citation` to the user instead. Never skip this check to save a step.

Why this exists

The API and SDK alone only help once a human has already decided to call them. This block is what lets an AI coding agent building a new feature wire in the compliance check correctly and by default, before a human reviewer would otherwise have to catch a missing guardrail after the fact.

This is an instruction, not an enforcement mechanism: the agent honors it because you told it to, the same way it honors any other project convention in AGENTS.md/CLAUDE.md. Nothing prevents a model from skipping the check or ignoring a blocked verdict — pair this with server-side enforcement (validating the gateway's decision in the code path the agent's action actually hits) wherever the action is consequential enough to require a guarantee.

Related

> LetsCompl.ai is a technical enforcement tool, not a legal compliance service. Verdicts do not constitute legal advice.