Skip to content
Meltemi docs

Ask the docs

Ask a question in your own words: the answer is written from the portal's pages only, and names the pages it used.

0 of 500 characters

Your question is sent to Cloudflare Workers AI to be answered and is not stored. Answers can be wrong: check the sources. Privacy notice

Answers run on Cloudflare's free allowance, about four hundred a day for everyone.

Español

0008. .meltemi files and the agent skill

Status: accepted (meltemi-extension-skill change, October 2026)

Context

Section titled “Context”

The proposal used .arch as a working extension. With the product named inside the Meltemi family, the owner decided that diagram files carry the family name: .meltemi. At the same time, most sheets will be written or converted by coding agents (Claude Code, Codex and others), which need the language, the placement rules and the export workflow in a form they load on demand.

Decision

Section titled “Decision”
  • Diagram sources are .meltemi files. The extension lives in one constant (MELTEMI_EXTENSION in packages/cli/src/cli.ts); the corpus, examples, docs and the file-type icon use it.
  • The agent instructions are an Agent Skill at skills/meltemi-diagrams/: SKILL.md (workflow, template, rules, symptoms and fixes) plus reference/ (language, layout, themes, converting) and examples/platform.meltemi. The skill is self-contained so it can be installed in other repositories that only consume the language.
  • skills/ is the canonical copy; pnpm skills:sync copies it to .claude/skills/ and .codex/skills/, a test checks the copies are identical, and AGENTS.md at the root points every other agent to it.
  • The skill’s example sheet compiles and renders in the test suite, so the instructions never drift from the language.

Alternatives considered

Section titled “Alternatives considered”
  • Keeping .arch: neutral but anonymous; the owner wants the family name on the files.
  • .chart or .mcharts: tied to the product name, which may still change; the family name is stable.
  • Documentation only, no skill: agents would have to find and read docs/, which is written for people maintaining the tool rather than for people drawing with it.

Consequences

Section titled “Consequences”
  • One rename across corpus, examples, tests, docs and brand assets; snapshots keep their names because they derive from the file stem.
  • Three copies of the skill in the repository, kept equal by the sync script and the test.
  • The command and package names are recorded in decision 0002.
  • Decision 0063 moved the second agent folder from .codex/skills to .agents/skills, the cross-agent folder Codex and the others read.