proofread

name	proofread
description	Proofread Markdown for spelling, grammar, typos, punctuation, and light style/clarity issues. Use for proofread, spell check, grammar check, typo review, or proofreading a Markdown file.
model	sonnet

Proofread skill

Adapted from Peter Hartree's HartreeWorks/skills repository.

Proofreading for Markdown documents using British English conventions by default. The scripts are for .md, .markdown, and .mdx files only.

Two engines are available:

• Spellcheck (fast, deterministic): Uses aspell for spell-checking (~2 seconds)
• LLM (thorough, AI-powered): Uses Gemini Flash for spelling, grammar, style, and clarity (~30-60 seconds per 100 lines)

Workflow

Step 1: Confirm target and mode

If the user did not provide a file path, ask for one. If the file is not Markdown (.md, .markdown, or .mdx), say this skill is scoped to Markdown files and do not run the bundled scripts.

Infer the engine when the user already gives a clear preference:

• Use Spellcheck for "quick spell check", "check spelling", or other spelling-only requests.
• Use LLM level 2 for generic "proofread", "check grammar", "review for typos", or "light style" requests.
• Use LLM level 1 for explicitly mechanical-only requests.
• Use LLM level 3 for comprehensive or thorough editing requests.

If the mode is unclear, ask:

Which proofreading approach do you want?

Spellcheck — Fast deterministic spell-check using aspell (~2 seconds). Good for quick checks. No API key required.

LLM — AI-powered proofreading using Gemini Flash (~30-60s per 100 lines). Choose a level:

• Level 1 — Mechanical only: Spelling, punctuation, grammar (fast, minimal output)
• Level 2 — Light style pass: Level 1 + top 5-10 style/clarity suggestions (recommended)
• Level 3 — Comprehensive: All style/clarity suggestions (thorough, more output)

Step 1a: Check API key (LLM modes only)

Skip this step if the user chose Spellcheck.

Resolve the current tool's skill directory before checking configuration:

• Codex: /Users/pablostafforini/My Drive/dotfiles/bin/agent-skill path proofread --tool codex
• Claude Code: /Users/pablostafforini/My Drive/dotfiles/bin/agent-skill path proofread --tool claude

Set skill_dir to the parent directory of the returned SKILL.md path. Use that directory for .env, scripts, and dependency installs.

If the user chose an LLM level, check if the API key is configured:

1. Read $skill_dir/.env
2. Check if it contains GOOGLE_AI_API_KEY= with a value

If the key is missing or the file doesn't exist, tell the user:

The LLM proofreading modes require a Google AI API key to use Gemini.

To set it up:

1. Go to https://aistudio.google.com/app/apikey
2. Create an API key
3. Add it to this skill's .env file:

   GOOGLE_AI_API_KEY=your_key_here
   

Would you like me to create the .env file for you once you have the key?

Alternatively, you can use the Spellcheck mode which doesn't require an API key.

Important: Do not search for the API key in the user's shell environment or other locations. Only use the key from the skill's .env file.

Step 2: Run the proofreading script

For spellcheck engine:

cd "$skill_dir" && yarn -s proofread "<file_path>" --engine spellcheck

For LLM engine:

cd "$skill_dir" && yarn -s proofread "<file_path>" --engine llm --level <1|2|3>

Add --language american only if the user explicitly requests American English.

The script outputs JSON to stdout. Parse it and present to the user.

Step 3: Present results

Format the output like this:

## Proofreading complete: <filename>

**Auto-applied (<count> corrections):**
- Line <n>: "<from>" → "<to>"
- ...

**Suggestions for review:**
- [S1] Line <n>: <description> (suggested: "<replacement>")
- [S2] Line <n>: <description>
- ...

Corrected file saved to: <filename>.proofread.md

**To accept suggestions**, type their IDs (e.g., "S1 S3") or "all", or "none" to skip.

Spellcheck mode does not auto-apply corrections; it reports all possible spelling fixes as suggestions for review.

Step 4: Apply accepted suggestions

When the user provides IDs:

cd "$skill_dir" && yarn -s apply "<file>.proofread.md" <S1 S2 ...>

Or if they say "all":

cd "$skill_dir" && yarn -s apply "<file>.proofread.md" all

Or if they say "none":

cd "$skill_dir" && yarn -s apply "<file>.proofread.md" none

Step 5: Confirm completion

Re-read the final file path if needed to confirm it exists and suggestion comments were removed.

Final file saved to: <filename>.final.md

Applied: S1, S3
Removed: S2, S4

First-time setup

If the user hasn't installed dependencies yet:

cd "$skill_dir" && yarn install

Configuration

The skill uses these environment variables from .env (LLM modes only):

• GOOGLE_AI_API_KEY: Google AI API key for Gemini (required for LLM modes)
• PROOFREAD_MODEL: Model ID (default: gemini-2.0-flash)

The spellcheck engine requires aspell to be installed:

brew install aspell

Output files

• <filename>.proofread.md: Auto-corrections applied, suggestions as HTML comments
• <filename>.final.md: After accepting/rejecting suggestions

Notes

• Uses British English conventions
• Preserves author's voice and technical terminology