Skip to content

Latest commit

 

History

History
65 lines (49 loc) · 4.4 KB

File metadata and controls

65 lines (49 loc) · 4.4 KB

Glossary

Canonical Skillware terms. Prefer the primary column in new docs. Synonyms are fine once per page for readability; link here on first use.

Skill file roles follow Skill anatomy (#326). Do not use retired Mind / Body / Conscience labels.

Inclusive language: inclusive-language.md. Issue: #252.

People and runtimes

Primary Means Do not use for this Keep when
Operator Person who installs Skillware, sets credentials, runs the CLI, and owns fork / commit / PR admin, generic “user” —
Contributor Person or supervised agent opening PRs on this repository operator (unless they also run the host) —
Supervised agent Cursor, Copilot, or similar working on this repo under an operator bot, AI contributor, host agent “human operator” when contrasting with the agent
Host agent LLM + tool loop that loads skills contributor, operator, calling agent —
End user Person the host agent is serving (for example “ask the user to confirm send”) operator Skill Directive copy; templates

user is not banned. It stays for LLM role: "user", Keep a Changelog user-visible, OS user-level config (~/.config), and end user in Directives.

Do not global-replace user → operator.

Skill artifacts

Primary Means Avoid
Skill ID Registry string category/name (folder path, manifest.name, CLI ID) “name” alone
Skill directory Folder skills/<category>/<name>/ “the skill” when the folder vs class is ambiguous
Skill bundle That directory on disk, or the dict returned by SkillLoader.load_skill() (bundle["class"], bundle["instructions"], …) “package” (conflicts with the PyPI package skillware)
PyPI package Installable skillware distribution calling a registry skill a package

Skill anatomy (roles)

Role File Answers
Contract manifest.yaml Typed I/O, constitution, issuer, requirements
Effect skill.py What runs in execute()
Directive instructions.md When and how the host agent should call the skill (skill context, not host persona)
Assurance test_skill.py Offline bundle tests
Presentation card.json Catalog / UI metadata
Corpus kb/, data/, … Bundled knowledge Effect reads
Reference schemas/, maps Machine-readable adjuncts to Contract
Interface SkillLoader adapters Host API tool schemas

Retired: Mind, Body, Conscience as role names. Filenames did not change.

Deprecated synonyms

Instead of Write
calling agent (LLM loading skills) host agent
Skill Package Standard Skill bundle standard (CONTRIBUTING)
user (person running Skillware) operator
dummy (placeholder data) mock, stub, placeholder, sample
master a domain cover / use a skill in a domain

Operator configuration

Primary Means Avoid
EVM operator config Writable evm.yaml plus skillware.core.evm_config merge helpers for chain IDs, rpc_env names, and token registry metadata shared by defi skills — see EVM operator config putting RPC URLs or private keys in YAML
Shared address book (operator address book) Writable addressbook.yaml under user config — cross-skill contact identity (mail recipients and defi transfer EOAs) via skillware.core.mail_config — see Address book operator config ERC-20 token contracts; orchestration chains:
public_0x Optional EVM wallet field on an address-book contact — counterparty EOA for defi skills such as defi/evm_tx_handler token/router/pool contract addresses (those belong in evm.yaml → tokens:)
Orchestration chain Named multi-skill pipeline under top-level chains: in .skillware.yaml (skillware chain run) EVM blockchain network
EVM chain (network) JSON-RPC network entry (ethereum, base, …) under evm: / evm.yaml orchestration chain