avoid-ai-writing-router

Route AI-writing audit, rewrite, edit, and verification requests across specialized skills.

5.0k|441|Updated Mar 6, 2026
Ready to run
npx skills add https://github.com/conorbronsdon/avoid-ai-writing --skill avoid-ai-writing-router-conorbronsdon
💬 Prompt for ChatGPT / Claude Chat▼
Please help me install this Agent Skill.
Skill: avoid-ai-writing-router
Source: https://github.com/conorbronsdon/avoid-ai-writing/tree/main/skills/avoid-ai-writing-router
Command: npx skills add https://github.com/conorbronsdon/avoid-ai-writing --skill avoid-ai-writing-router-conorbronsdon

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Requests involving AI-writing cleanup often mix detection, rewriting, file editing, and verification in one prompt, and no single narrow skill owns the whole job. This router classifies the request once, picks the correct specialized skill, and carries context between stages so nothing is re-inferred or lost. ## Core Features & Use Cases - Intent routing: Sends detect-only requests to the detector, returned-text rewrites to the rewriter, named-file edits to the file editor, before/after comparisons to the verifier, and consequential authorship claims to the false-positive reviewer. - Handoff envelope: Passes a compact YAML envelope (intent, source kind, protected constraints, execution evidence, pass budget) between skills so downstream stages reuse prior decisions. - Bounded loops: Enforces a shared editing-pass budget (max 2), one repair re-entry, and explicit stop conditions to prevent infinite detect-repair cycles. - Graph validation: Ships a stdlib-only Python validator that fails closed on dangling edges, unbounded cycles, terminal-node leaks, or drift between the JSON graph and its documentation. - Use Case: A user asks to "scan this post, rewrite it, and confirm nothing important changed." The router sequences detector → rewriter → verifier, tracks the two-pass budget, and reports which stages ran and the final preservation status. ## Quick Start Route this request through the right Avoid AI Writing workflow: audit this paragraph for AI-writing patterns, rewrite it in my voice, and verify the rewrite preserved all facts.

Frequently Asked Questions about avoid-ai-writing-router

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I route a multi-step AI-writing cleanup request?▼

Invoke the router with the full request, such as scan, rewrite, and verify. It classifies intent once, builds a handoff envelope, and sequences the detector, rewriter or file editor, and preservation verifier within a shared two-pass editing budget.

What is the handoff envelope in skill orchestration?▼

The handoff envelope is a compact YAML structure carrying intent, source kind, protected constraints, execution evidence, detector summary, and pass budget between skills. It prevents each downstream stage from re-inferring context and forbids marking execution fields as executed without host evidence.

Can the router edit a file directly?▼

No. The router only selects owners and sequences stages. Named-file mutation belongs to file-edit-in-place, returned-text rewriting to voice-preserving-rewriter, and verification to preservation-verifier, each with explicit ownership rules.

How are repair loops bounded in the skill graph?▼

All mutations share one editing-pass budget of at most two passes. A verifier FAIL may re-enter the correct repair owner once via a REPAIR edge with max_reentries set to 1, followed by one recheck; exhausted budgets are reported as unresolved failures.

When are detector results escalated to false-positive review?▼

Escalation happens when detector signals are used for consequential claims like proving AI use, cheating, or hiring decisions. The false-positive-reviewer is terminal and returns control to the router if fresh signal collection is needed.