Crate hatchling_skill 

Signed skills: the trust core.

Part of Hatchling, published so hatchling-runtime can be installed. It carries no stability promise on its own.

Every skill is signed, sandboxed, and capability-scoped. An unsigned skill, a tampered skill, or a skill asking for capabilities it wasn’t granted does not run — fail-closed, structurally, not by convention.

This crate implements the signing + manifest + admission trust core. The broker verifies signatures before admission; the cell never self-attests.

Structs

Admission

A successfully admitted skill: the verified manifest plus the capabilities the broker will enforce at call time.

CodeHash

The SHA-256 hash of a skill’s code, pinned at admission.

Manifest

A skill manifest without the signature: the signed payload.

RegisteredSkill

A skill admitted to the registry: the verified admission plus the pinned code hash.

SignedManifest

A manifest with its Ed25519 signature.

SkillId

A unique identifier for a registered skill.

SkillRegistry

The broker’s registry of admitted skills.

Enums

Error

All ways a skill can be refused.

Constants

SIGNING_CONTEXT

Domain tag that starts every signed manifest payload (format v1).

Functions

check_trust_root

Checks that root can serve as a trust root: a valid Ed25519 point, and not a weak (small-order) key, under which a signature can verify for any message.

verify_admission

Verify a signed manifest and admit it against the user’s grants.