fmueller/scribae
CLI to turn Markdown notes into SEO briefs, drafts, metadata, and translations using LLMs.
What's novel
CLI to turn Markdown notes into SEO briefs, drafts, metadata, and translations using LLMs.
Code Analysis
10 files read · 3 roundsA CLI tool that orchestrates a multi-stage LLM workflow (idea → brief → write → feedback → refine → meta → translate) to turn local Markdown notes into structured SEO content packages, with a novel offline MT + LLM post-edit translation pipeline that preserves Markdown structure.
Strengths
The translation pipeline is genuinely well-engineered: markdown-aware segmentation, protected token substitution, batched seq2seq MT with model routing (Marian/NLLB/pivot), LLM post-editing with strict structural validation, and a multi-level fallback chain. The codebase demonstrates consistent patterns across all workflow stages with typed Pydantic outputs, language enforcement, and clean separation between CLI, business logic, and prompts.
Weaknesses
The core LLM workflow stages (idea, brief, write, feedback, refine) are largely prompt engineering + pydantic-ai wrappers rather than novel algorithms — the real substance is in the translation pipeline. There's significant code duplication across the workflow modules (each repeats the same prepare_context → generate → render pattern with near-identical error handling), and the asyncio.run-in-sync pattern for LLM calls is a code smell that limits testability.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
131
Contributors
2
Files
83
Active weeks
13
Repository
Language
Python
Stars
1
Forks
0
License
Apache-2.0