Marketplace and SDK

Every marketplace package includes an AGENTS.md file that bootstraps agent context. The manifest defines capabilities, required permissions, input/output schemas, and side effect declarations.

The manifest is generated automatically during the GitHub import pipeline and can be customized before publishing.

Tools are defined with explicit input schemas, output schemas, side effect declarations, and risk tiers (read/write/danger). The tool policy matrix governs per-tool behavior.

Each tool declaration includes max calls per run, approval thresholds, offline behavior, cost weight, and the roles allowed to invoke it.

The Productivity canvas supports drag-and-drop composition of intelligence blocks. Connect CRM agents, email drafters, research pipelines, and custom workflows visually.

Productivity actions produce canonical receipts and bind to work items. Pipeline records and comms are first-class objects in the Productivity CRM foundation.

The GitHub import pipeline provides secure repository cloning, security scanning, file indexing, and AGENTS.md generation. Every imported repository undergoes security analysis before marketplace listing.

Import state is tied to Nome policy — tenant guardrails, visibility flags, and install rules are enforced at the platform level.

Marketplace publishing follows a scan → index → review → publish pipeline. Security scanning is mandatory. Published packages receive tenant-scoped install state and rollout tracking.

The builder workspace and marketplace install/update/disable paths mint canonical work items and refs.

Marketplace imports and packages stay behind tenant guardrails, security scanning, and explicit installability rules. Install state is Nome-owned — not managed by a separate vendor control plane.

The guardrail test suites validate that marketplace operations respect tenant boundaries, visibility policies, and billing prerequisites.