Tried sponsors/conorbronsdon: A Practical AI-Writing Cleanup Skill
conorbronsdon is gaining attention quickly—+179 stars today—because it addresses a problem most AI developers eventually hit: content that is technically correct but immediately recognizable as machine-written.
The skill audits generated text for common AI patterns—repetitive transitions, over-structured sections, inflated claims, generic conclusions, and unnatural phrasing—then rewrites the content to sound more direct and human. It is designed to work with agents you already use, including Claude Code, OpenClaw, Codex, and Hermes, rather than forcing another writing interface into your workflow.
The useful pattern is to run it as a final review step:
draft -> technical review -> AI-pattern audit -> final editor
For teams using a custom Anthropic-compatible gateway, the connection can stay simple. For example, with B-Lost Universal Relay:
export ANTHROPIC_BASE_URL="https://b-lost.com/v1"
export ANTHROPIC_API_KEY="$B_LOST_API_KEY"
claude \
--model claude-fable-5 \
"Audit this draft for AI writing patterns, then rewrite it without changing technical claims."
The exact environment variable names depend on the client, but the important part is using an Anthropic-compatible /v1/messages endpoint. The same architecture works with Cursor, Cline/Roo Code, Windsurf, Aider, LibreChat, and other clients that support custom API bases.
One cost detail matters for repeated editing loops: B-Lost supports native Anthropic prompt caching, with 90% discounts on cache hits. That is useful when the agent repeatedly receives the same style guide, repository context, or editorial rules.
B-Lost also lists 20% off official pricing and a 100% first-deposit bonus for deposits of at least $10. I would still benchmark latency, model behavior, and rate limits before routing production traffic through any relay.
My quick take: this skill is most valuable as a deterministic “last mile” in content agents—not as a replacement for human editing, but as a fast way to remove the obvious fingerprints before review.
Top comments (0)