InKCre Agent Skills

Organization-level common instructions and reusable execution assets for InKCre coding Agents.

Scope

This repository owns a reusable Agent Skill only when at least two repositories need the same executable guidance and keeping separate copies would create drift. Repository-local AGENTS.md, commands, architecture, deployment behavior, and product-package skills remain with their code.

Current external owners are recorded in catalog.json:

• edit-svc-shared-docs stays in InKCre/docs because its workflow and check script operate on that Hub.
• ui-web stays in the published @inkcre/ui-web package because it describes that package's components and ships with the product artifact.

Consumption

Common instructions

Clone this repository at a reviewed commit, then install the Codex user profile:

./install.sh
./install.sh --check

The installer links global/AGENTS.md to ~/.codex/AGENTS.md. It refuses to replace an existing path. The instructions therefore apply to every Codex project for that user and must remain safe outside an InKCre checkout. Repository-specific rules stay in each repository's own AGENTS.md.

Update by reviewing and checking out a newer repository commit. Roll back by checking out the previous commit. Restart Codex after either operation.

Skills

Portable skills under skills/ can be inspected or installed with the standard open Agent Skills CLI:

npx skills add InKCre/agent-skills --list
npx skills add InKCre/agent-skills --skill <name> --agent codex
npx skills add InKCre/agent-skills --skill <name> --agent codex --global

Project scope is the default. Use global scope only when the skill applies to work outside one repository. In automation or another reproducible environment, use a reviewed commit in the direct skill URL:

npx skills add https://github.com/InKCre/agent-skills/tree/<commit>/skills/<name> --agent codex

The repository currently has no admitted shared Skill, so --list reports no skills until the first asset proves two consumers.

For an asset owned outside this repository, use the smallest mechanism supported by the consumer:

1. load an installed or package-delivered skill directly;
2. add a thin local discovery wrapper that points to the canonical owner;
3. add a versioned shared mount when the wrapper must read the canonical asset.

There is no synchronization bot or cross-repository writer. Consumers update their own references through ordinary reviewed changes.

Layout

global/AGENTS.md          user-scope common coding Agent instructions
skills/                  canonical skills owned here, when admitted
templates/AGENTS.root.md starting point for a new repository
catalog.json             owner and consumer discovery metadata
tasks/                   active task packets

The consumption split follows the useful part of arcboxlabs/agent-skills: keep global instructions separate from portable and project-scoped skills, and make installation idempotent and conflict-preserving. InKCre does not carry its multi-client generation, vendored skills, checks, or Git hooks without a proven consumer.