Yoker Writing Assistant

A writing assistant that coaches and develops — but never writes for you.

This is a Yoker showcase package. It demonstrates the yoker-as-runtime mode: Yoker is the entry point, and this package provides the main agent and skills for an interactive writing coaching session. The agent interviews, challenges, reviews, and flags — but every word of prose is written by the author.

Status

Initial release. The Python package, plugin manifest, entry point, agent definition, and all 8 skills are in place. Tests pass, all quality gates green. See AGENTS.md for the full project guide.

What It Does

The writing assistant is a developmental editor + writing coach for an author who writes every word themselves. It operates on the declared-work scope: it only acts on prose the author has already written. It never generates prose, never rewrites, never fills blanks.

Skills

The Non-Negotiable Rule

The agent never writes prose. Every gap is marked TODO:, every proposal is marked TODO PROPOSAL:. The author accepts, rejects, or writes. This is the core design principle — friction is the value, not a bug.

Quick Start

uvx yoker-writing-assistant                    # run directly (if published)

From a local checkout:

make env-dev                                    # install all dependencies
make run                                        # launch the writing assistant

Or run directly via the Yoker CLI:

yoker --with yoker_writing_assistant --agent yoker_writing_assistant:writing-assistant

A Yoker backend is a prerequisite — either a local Ollama install or a cloud LLM provider API key. If you do not already have one, run uv run yoker init once to write ~/.yoker.toml with a backend of your choice.

Configuration

Runtime configuration lives in ~/.yoker.toml (never committed). A reference template is provided as yoker.toml in this repo. The key sections:

• [backend] — LLM provider and model
• [agents.directories] — directories to scan for agent definitions (e.g., c3 = "../c3/agents" for the optional researcher agent)
• [skills.directories] — directories to scan for skill definitions (e.g., c3 = "../c3/skills" for c3 heritage skills)
• [plugins] — plugin registration (this package is loaded via --with yoker_writing_assistant)

Optional: c3 Researcher

The writing assistant can delegate research tasks to c3:researcher, an external agent from the c3 project. To enable it, add to yoker.toml:

[agents.directories]
c3 = "../c3/agents"

[skills.directories]
c3 = "../c3/skills"

If c3 is not configured, research delegation is unavailable and the agent falls back to direct web search (yoker:websearch/yoker:webfetch), noting the limitation in its output. All other functionality works without c3.

Architecture

This package is a Yoker plugin: it declares an __YOKER_MANIFEST__ that points Yoker to its agents/ and skills/ directories (inside the Python package, discovered via importlib.resources). Yoker's plugin loader discovers the agent and skill definitions automatically, namespacing them under yoker_writing_assistant:. The entry point is a thin Python wrapper (cli.py) that injects --with yoker_writing_assistant and --agent yoker_writing_assistant:writing-assistant into Yoker's CLI, so uvx yoker-writing-assistant launches the writing assistant directly.

The sister project yoker-assistant demonstrates the complementary mode — yoker-as-SDK — where Python owns the process and calls Yoker as a library.

Both projects share a common quality bar documented in STANDARDS.md.

Documentation

Full documentation will live in docs/ (pending). The AGENTS.md file provides the project guide for agents working on this codebase.

License

MIT