External playbook packages are authored outside Tessera and executed inside Tessera. This keeps business-domain logic portable while preserving one governed runtime.

​
Required Surface

A package must expose:
  • Playbook graph definition.
  • Prompt references.
  • Script references when scripts are used.
  • Schema references for agent outputs and structured artifacts.
  • Declared artifacts.
  • Declared capabilities.
  • Fixture and golden inputs for deterministic package-local tests.
  • Final materialization targets for Markdown, CSV, JSON, or PDF outputs.

​
File Reference Rules

Use package-relative references. Do not use absolute paths, parent-directory escapes, or symlinks that leave the package. Do not reference generated files that cannot be recreated or validated.

​
Agent Output Rules

Every agent output must declare a schema. Prompts should tell agents to produce schema-conformant output. Review and rework loops must return to clear downstream nodes, and human review actions need explicit approved and change-requested paths.

​
Artifact Rules

Every final artifact must be materializable. V1 materialization targets are Markdown, CSV, JSON, and PDF. Final artifacts should name their audience and acceptance check. Use first-class effect nodes for materialization. Workspace outputs declare input.target.kind: "workspace", a package-relative path, and a format of markdown, json, csv, or pdf. Connector-backed outputs declare input.target.kind: "external" and rely on Tessera effect evidence for the final output reference. CSV outputs are UTF-8 comma-separated rows. Fields with commas, quotes, or newlines are quoted, and embedded quotes are doubled.

​
Capability Rules

Every tool or connector use must be declared. Prefer generic capabilities such as Gmail, web search, web fetch, or package-local deterministic scripts. Domain-specific connector semantics should stay inside the external package until promoted as a generic Tessera primitive.