mirror of
https://github.com/pandoscope/skills.git
synced 2026-09-19 05:58:34 +08:00
Claude
57d7d62515
A skill is installed individually, into a repo that may not use disambiguate or any other tool this repo assumes. The unit of self-containment is the skill's folder: everything in it ships together (derived/tdd/ brings tests.md and lint-red.sh with its SKILL.md), and nothing outside it ships alongside. Three consequences, all stated in the section: no glossary term references — any vocabulary, since the reason a link is wrong in a skill has nothing to do with whose term it is; no tracker CLI names, which was already here from the AGENTS.md resolution and is an instance of the rule rather than its own; and skill-to-skill references are name-only and must degrade, because `npx skills` has no dependency management today — revisit if the installer ever ships it. Teaching a format is explicitly not referencing a term — domain-modeling shows a link as an example of what a reader writes in their own glossary, and that stays. The rule is stated without reaching into another repo: explaining it by contrast with how the template distributes would make a convention about self-containment depend on knowing another repo's rules. The template's equivalent is tracked as frankify-app/agentic-engineering-template#72. No violations today: no file under original/ or derived/ links a glossary term. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013bmY2VqQ4S6gR8NJ7TjW1G