Learning Center Spec

How the agent learning center should work

The learning center is not marketing copy. It is the machine-friendly documentation layer that lets agents discover tools, read schemas, understand failures, and compose recipes safely.

Open Learn Read connect guide

Why the learning center exists

• If agents only see scattered APIs, they cannot reliably discover which tools are available.
• When input and output formats are inconsistent, agents are forced to guess and fail more often.
• Without manifests and changelogs, version upgrades break existing workflows silently.
• Without structured errors, agents cannot retry or degrade safely.

Scope

• Tool summary: what it does, when to use it, and its limits.
• Input and output schemas.
• Request and response examples.
• Error codes and retry strategy.
• Cost rules and resource usage.
• Quality checks and version history.

Recommended structure

/learn: learning center home
/learn/tools: tool catalog
/learn/tools/{tool_id}: tool detail
/learn/tools/{tool_id}/versions/{version}: version detail
/learn/recipes: orchestration patterns
/learn/changelog: global change log

Machine-readable endpoints

GET /api/tools/catalog: tool catalog with version, status, scope, and cost.
GET /api/tools/{tool_id}/manifest: tool manifest.
GET /api/tools/{tool_id}/examples: request and response examples.
GET /api/tools/{tool_id}/errors: structured errors and handling guidance.
GET /api/learn/recipes: reusable workflow templates.

Minimum manifest fields

• tool_id, version, summary
• input_schema, output_schema
• required_scopes, cost_rule
• latency_sla_ms, error_codes, quality_checks

Rollout priority

1. Start with the machine-readable catalog and manifests.
2. Then ship detail pages and examples.
3. Finish with recipes and automated changelog generation.