matchplane-agent-integration

Connect buyer and seller agents to MatchPlane's bounded Agent, Skill, MCP, routing, and retrieval contracts. Use when adding an external agent, designing buyer/seller handoff, exposing MCP tools, or reviewing token ownership and contact-consent boundaries.

Jump to install

Repository: LIghtJUNction/matchplane
Last source activity: August 23, 2026 at 05:40
Detected SKILL.md language: English
Stars: 3
Forks: 4

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.

A direct command skips the review prompt. Inspect the source before running it.

npx skills add https://github.com/LIghtJUNction/matchplane --skill matchplane-agent-integration

The command stays on one line. Scroll horizontally to inspect it before copying.

Prefer a local copy? Download the files currently available to SkillsMP.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

File Explorer

5 files

SKILL.md

Source instructions · Read-only preview

Back to search Back to installation

MatchPlane Agent Integration

Use the same integration shape for buyers and sellers. The role changes the intent and allowed tools; identity, platform scope, audit, and consent rules do not change.

Integration workflow

Declare the current store path, party side, and a domain-neutral narrative. Never put category-specific fields in the mall request unless the selected store owns that schema.

Authenticate with a Better Auth session for a human chat or an organization API key for a machine agent. Keep the key server-side and request only the smallest platform:read, retrieval:query, or marketplace permission needed.

Call the mall route boundary once. Treat selected stores, source, and degraded as advisory routing evidence, not as permission to disclose contact details or settle money.

At each selected store, invoke only the MCP tools advertised by its immutable manifest. The store owns its catalogue, vector store, embeddings, and ranking Skill; stores never route to other stores.

Return canonical asset/merchant references and explainable reasons. Ask the root to verify active listing state, seller authorization, exposure billing, and consent before introducing parties.

Release WeChat/phone/contact data only through the platform contact-consent flow after both parties opt in. A successful AI match is never consent.

Token and failure policy

External buyer/seller agents pay for their own model calls. Mall-initiated selection has bounded candidate, input/output, rate, and timeout limits with an auditable cost_bearer: "platform" ledger. Do not retry a degraded response in a loop; surface it and let the caller decide whether to try another provider.

Use references/agent-envelope.md for request/result fields and the minimal MCP tool contract. The normative JSON schemas remain in docs/agent-mcp-skill-protocol-v1.json, docs/platform-routing-protocol-v1.json, and docs/retrieval-protocol-v1.json.