MAGIC Agent Skills is now open source! Star on GitHub
MAGIC Agent SkillsMAGIC Agent Skills
Commands

/linguistic:spec

Create or edit a structured workspace spec document under specs/.

Syntax

/linguistic:spec [spec-type]

Description

Create or edit a structured spec document for the current workspace under specs/. Common spec types:

SpecOwner SkillPurpose
language-spec.mdlinguistic-scopeLanguage identity, typology, resource class, vitality
data-spec.mdlinguistic-corpus + bitextData manifest with sources, licenses, sizes, dedup stats
tokenizer-spec.mdlinguistic-tokenizeTokenizer config, fertility audit, vocab extension plan
eval-spec.mdlinguistic-evalBenchmark choice, metric choice, contamination assessment
model-card.mdlinguistic-ethicsPublic-facing model card with uses, limits, attribution

Specs are user-editable and reviewed before phase advancement.

Options

ArgumentDescription
[spec-type]Type of spec to create (language, data, tokenizer, eval, model-card)

Example

/linguistic:spec tokenizer
> Creating tokenizer-spec.md for Yoruba (yor)...
Writing to workspace/specs/tokenizer-spec.md

# Tokenizer Spec: Yoruba
- Current tokenizer: tiktoken-cl100k_base
- Fertility: 3.4× (EXTEND MANDATORY)
- Method: OFA vocab extension
- Config: model_type=unigram, vocab_size=32K, byte_fallback=true
...
  • /linguistic:review — review all specs in context
  • /linguistic:propose — plan generates specs automatically
Was this page helpful?
Edit on GitHub

Last updated on

On this page