Instruction file imported from rhein1/agoragentic-integrations (
.cursor/rules/agoragentic-integrate.mdc). Copyright stays with the author.
Agoragentic Integrate
Prefer thin adapters over forks or duplicate runtimes.
For each external system:
- Identify native lifecycle/tool hooks and execution boundaries.
- Map them into existing Harness Core middleware events.
- Reuse existing policy and receipt schemas rather than duplicating them.
- Store hashes, bounded summaries, status, and evidence refs instead of raw private outputs by default.
- Keep local integration separate from hosted activation, marketplace publication, payment enablement, or trust mutation.
- Add deterministic fixtures and one demonstrable user outcome before advertising support.
Useful integration targets include coding-agent hosts, web/data engines, quality/security scanners, optimization systems, and rendering engines.
For Anthropic Commerce Agents, use the existing claude-agent-sdk/commerce/
title-edit fixture and its pinned shared-executor conformance path. Keep host
approval outside model tools and revalidate the exact change at apply time.
Read its qualification record before claiming support: SDK control-wire tests
with synthetic CLI peers do not establish real CLI enforcement or model-runtime
compatibility. Never repoint the synthetic store at a live merchant.
Advanced Context
- Commerce fixture and qualification: https://github.com/rhein1/agoragentic-integrations/tree/main/claude-agent-sdk/commerce
- agent workflow contracts: https://github.com/rhein1/agoragentic-integrations/blob/main/docs/agent-workflow-contracts.md
- MCP adapter: https://github.com/rhein1/agoragentic-integrations/tree/main/mcp
- federation and interchange: https://github.com/rhein1/agoragentic-integrations/tree/main/interchange