<< All versions

Skill v1.0.0

currentLLM-judged scan95/100
wadackel/dotfiles/vocab
──Details
PublishedOctober 23, 2026 at 02:37 PM
Content Hashsha256:d6be09d38535001a...
Git SHA78b96dce0dd9
──Files
Files (1 file, 3.9 KB)
SKILL.md3.9 KBactive
SKILL.md · 57 lines · 3.9 KB

version: "1.0.0" name: vocab description: >- Registers and maintains the owner's vocabulary: the short words and workflow names they use in instructions, kept in the Obsidian vault's 06_Vocabulary/ and injected into every Claude Code, Codex, opencode, and Hermes !claude session. Use when the owner defines a term or asks to remember what a word means ("この語を登録して", "語彙に追加", "〜は〜のこと、覚えておいて", "register this term", "add to vocabulary"), or asks to check or apply the vocabulary ("語彙を反映して", "語彙をチェックして", "vocab apply", "vocab lint"). Not for knowledge notes (use llm-wiki) or for Claude Code auto memory. argument-hint: "[add <term> | apply | lint | weekly]"


vocab

The vocabulary lives in the vault, not in this repository. Every change goes through ~/.agents/scripts/vocab.ts; never edit 06_Vocabulary/ or 98_Maintenance/proposals/Vocabulary/ with file tools.

Add or change a term

Every addition and change is a pending proposal that becomes vocabulary only when the owner sets status: approved in the 06_Vocabulary/語彙レビュー Bases view. Nothing, including a definition the owner dictated, is written to 06_Vocabulary/ directly: this command runs without a permission prompt, so text injected into a session must not be able to become approved vocabulary.

bash
~/.agents/scripts/vocab.ts add <term> [--kind <term|workflow|workflow-step|decision|repo>] [--definition "<one or two sentences>"] [--alias <a>]... [--rel <relation>=<term>]... [--refers-to <path>]... [--path <repo path>] [--draft] [--origin explicit|weekly]
  • A new term needs --kind and --definition. For an existing term the same command proposes a change: a new definition, added relations, or added aliases
  • Add --draft when you wrote the definition rather than the owner; the proposal's evidence line says which
  • Keep the definition to what the owner said; do not fill gaps from general knowledge
  • Relations allowed by 06_Vocabulary/_schema.md: distinct_from (a term it is confused with), applies_in (a repo note), part_of (a workflow), supersedes, and for decisions because / rejected
  • A term used only in one repository gets --rel applies_in=<repo>; a repo term gets --path so a relative refers_to can be checked

Report the proposal path it printed and that it waits for the owner's approval.

Weekly drafting

The memo scripts only record material — corrections, answers to your questions, and unregistered terms the owner repeats — in ~/.local/state/vocab/; no model writes a definition at session end. Once a week (from /weekly-review, or when the owner asks), turn that material into proposals:

bash
~/.agents/scripts/vocab.ts propose --weekly
  1. It writes the proposals that need no judgment (a refers_to that no longer exists, an owner-invoked skill with no term) and prints a packet: the week's excerpts, recurring terms, and the "Alternatives Considered" of this week's plans, plus how many more automatic proposals fit (at most five wait at a time).
  2. For each item in the packet that shows the owner using a word in their own sense, draft one proposal with add --draft --origin weekly (it refuses once five automatic proposals are waiting). The definition says only what the excerpt shows; a term whose meaning the packet does not show is skipped, not guessed. A correction usually also means distinct_from the term that was misread.
  3. Stop when the packet's remaining count reaches zero.
  4. Reflect the owner's approvals and check the result:
bash
~/.agents/scripts/vocab.ts apply
~/.agents/scripts/vocab.ts lint

Apply and lint on request

apply turns proposals the owner set to approved into notes and archives rejected ones; lint exits 1 on errors. Run them outside the weekly flow only when the owner asks. Report lint errors verbatim; fix a note only through an add proposal.

All versions