Skip to main content

tmp-schema

Generate a complete waz Token Model Protocol (TMP) schema from a CLI's help text. Use when creating or regenerating ~/.config/waz/schemas/<tool>.json for the command palette, waz tmp, or waz resolve.

Informations de source

Dépôt
codeitlikemiley/waz
Dernière activité de la source
20 août 2026 à 13:27
Langue détectée de SKILL.md
anglais
Étoiles
14
Forks
1

Options d'installation

Le prompt qui vérifie d'abord la source est sélectionné par défaut. Vous pouvez passer à une commande directe ou télécharger une copie locale.

Vérifiez les fichiers source

Lisez SKILL.md et les fichiers associés affichés par SkillsMP avant de décider de l'installer.

Explorateur de fichiers
2 fichiers

Affichage de SKILL.md

SKILL.md
Instructions source · Aperçu en lecture seule
name
tmp-schema
description
Generate a complete waz Token Model Protocol (TMP) schema from a CLI's help text. Use when creating or regenerating ~/.config/waz/schemas/<tool>.json for the command palette, waz tmp, or waz resolve.
# TMP schema generation You convert CLI `--help` into a **complete** JSON array of command entries for waz. Read `references/tmp-contract.md` before emitting JSON. ## Goal Cover **every subcommand and nested command** present in the help harvest, not a popular subset. Prefer completeness over brevity. If help is truncated, still emit every command named in the harvest and mark missing flags only when the help for that command was not provided. ## Output Return **only** a JSON array (no markdown fences, no commentary). Each element: ```json { "command": "tool sub", "description": "one line", "group": "tool", "tokens": [] } ``` ## Rules 1. `command` is **only** the binary plus subcommand words (`git commit`, `docker compose up`). Never put `[args]`, `<file>`, or flags in `command`. 2. Include a base entry `command: "<tool>"` for flags that apply with no subcommand. 3. One JSON object per invocable command path. Nested CLIs (`docker compose logs`) are separate entries, not flattened into `docker`. 4. Every flag and positional in that command's help becomes a token. 5. Token `default` is always a string or JSON null (`"false"`, `"0"`, never a JSON boolean/number). 6. Token `flag` is `"--verbose"`, `"-n"`, or JSON null for positionals. Never `false`. 7. `token_type`: `Boolean` for switches, `Enum` when help lists choices (`values: [...]`), `File` for paths, `Number` for ints/floats, else `String`. 8. Do not invent subcommands that are not in the harvest. Do not skip ones that are. 9. Prefer built-in resolvers over empty enums when the value is dynamic: - git files → `{"resolver":"git:status_files"}` - git branches → `{"resolver":"git:branches"}` - git remotes → `{"resolver":"git:remotes"}` - cargo bins/examples/packages/features/profiles/tests/benches → `cargo:<name>` - npm scripts → `{"resolver":"npm:scripts"}` For brew installed packages use `{"command":"brew list -1","parse":"lines"}`. Never `brew formulae` or `brew casks` (too large; hits the data_source timeout). Otherwise use `"data_source": {"command": "<safe short list cmd>", "parse": "lines"}`. 10. Boolean flags default to `"false"` and `required: false`. 11. Required positionals: `required: true`, `flag: null`. 12. `group` is the tool name (the binary). 13. Repeatable positionals/flags (`git add` paths): `"repeat": true` so multiple `--set path=` values append. ## Completeness checklist - [ ] Base command present - [ ] Every harvested `=== tool … --help ===` heading has a matching `command` - [ ] Nested commands included - [ ] Help/version meta commands omitted - [ ] Flags are tokens, never commands
Voir sur GitHub