Implementing skill manifests

This document describes the practical pieces for generating, validating, and consuming skill manifests in the agential semioverse.

Specification

• JSON Schema: library/mathematics/objects/agential-semioverse/tools/skill-manifest/specs/skill-spec-v1.schema.json.
• Source SKILL files conform to Skill Spec v1.
• Only entries with a runtime are considered executable; others are advisory.

Tooling

• Build: python library/mathematics/objects/agential-semioverse/tools/skill-manifest/scripts/build-skill-manifest.py --output library/skill-manifest.json
• Lint: python library/mathematics/objects/agential-semioverse/tools/skill-manifest/scripts/lint-skills.py
• Manifest path: library/skill-manifest.json (consumed by engine API and MCP).

Integration points

• Engine startup rebuilds the manifest if missing and validates for duplicate ids and runtimes.
• MCP/OpenAPI surfaces only skills present in the manifest and allowed for the requesting agent.
• Tests should cover: manifest generation, missing runtime rejection, duplicate id rejection.

Future work

• Add manifest fragments for per-agent scopes.
• Emit reference URIs to encyclopedia concepts for tokens in inputs/outputs.