ai-writing-suite
surahli123/ai-writing-suite
Open-source AI writing-assistant skill suite: polish/de-AI prose, learn an author's voice, and (v2) KB-guided Q&A + drafting. Engine + pluggable knowledge base; ships Claude & Codex packaging.
Instalação
npx convez-ai surahli123/ai-writing-suite1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
README
A tradução para o português deste README ainda está na fila.
AI Writing Suite
An open-source, MIT-licensed agent skill suite for writing assistance — the public version of a writing-assistant skillset for a company Data Science team. It polishes, drafts, and reviews prose against a pluggable knowledge base (a "DS Comms Playbook" in the company fork; a generic best-practices KB in this OSS build).
Engine vs fuel: this repo is the engine. The knowledge base is fuel — swappable. The OSS build ships a generic KB; a company fork drops its real playbook into the same slot. The proprietary playbook never enters this public repo.
What it does (target)
- Polish / review — de-AI and tighten prose while preserving meaning and author voice.
- Voice learning — interview + distill a writer's historical style into a reusable profile.
- Knowledge QA (mini-RAG) — answer questions over the playbook + best practices (v2).
- Guided drafting — draft a page using the playbook (v2).
Plus a human-gated self-improvement loop: each session can propose new rules; you approve before anything is written. Core skill logic is never auto-edited.
Status
v1 — built, verified, and installable on Claude, Codex, and Cursor. Ships the polish/de-AI
capability (comms-polish), voice learning (voice-onboard), a generic pluggable KB seed with an
end-to-end retrieval smoke path, an eval harness (Python AI-tell detector + fixtures), and a
human-gated self-improvement hook. comms-qa (KB Q&A) and comms-draft are v2 stubs.
- Plan (read first):
docs/design-ai-writing-suite-v1-2026-06-06.md— decision log D1–D12, v1 scope, risks, eng+ceo review report.
Install
One source of truth (skills/ai-writing-suite/) published to three hosts.
Claude Code
claude plugin marketplace add surahli123/ai-writing-suite
claude plugin install ai-writing-suite@ai-writing-suite
Claude auto-discovers the sub-skills (comms-polish, voice-onboard, …); invoke by asking it to
polish/review prose, or call a skill by name.
Codex
codex plugin marketplace add surahli123/ai-writing-suite
codex plugin add ai-writing-suite@ai-writing-suite
Codex resolves the plugin from the same source tree — no --sparse needed; the Codex
marketplace manifest sits at the repo root (.agents/plugins/marketplace.json).
Cursor
Cursor reads Anthropic-format Agent Skills from .cursor/skills/ — there is no marketplace, so
copy the skill tree into a discovery location:
git clone https://github.com/surahli123/ai-writing-suite
cp -R ai-writing-suite/skills/ai-writing-suite ~/.cursor/skills/ai-writing-suite
(Or copy into your project's .cursor/skills/ for project-scoped use.) Cursor discovers the skills
on the next session; invoke with /comms-polish, /voice-onboard, … or by intent.
Repo Layout
.claude-plugin/marketplace.json # Claude marketplace → points at the source tree
.agents/plugins/marketplace.json # Codex marketplace → points at the source tree
skills/
ai-writing-suite/ # THE SUITE — single source of truth
# (Claude + Codex install this; Cursor copies it)
SKILL.md # thin router
.claude-plugin/plugin.json # Claude plugin manifest
.codex-plugin/plugin.json # Codex plugin manifest
skills/ # sub-skills: comms-polish, voice-onboard, comms-qa*, comms-draft*
_shared/ # patterns (engine) + knowledge (swappable KB) + voice profile
evals/ # AI-tell detector + fixtures + smoke tests
NOTICE.md LICENSE README.md CHANGELOG.md
docs/ # design docs, handovers, plans, packaging (* = v2 stub)
Attribution
Absorbed material from other repos is MIT-licensed; each source's copyright + license is
preserved and credited in skills/ai-writing-suite/NOTICE.md.