email-issue-fixer
Helps your agent fix small errors in texts. It improves writing without changing the original style and removes tracking links.
What it does
This agent fixes common typos and grammar mistakes in your emails, like repeated words or incorrect subject-verb agreement. It can also remove tracking codes from links if you ask.
When to use it
- ✓Before sending an important email.
- ✓When you want to ensure your text is clear and error-free.
- ✓If you need to clean tracking links in an email.
Example of what to ask your AI
“Fix this email draft and remove tracking codes from the links.”
Want more? Open the sections below.
How to use
The easiest way: copy the message below and paste it into Claude, ChatGPT, Cursor or any AI. It reads the skill and follows the instructions.
Skill link
Ready-to-paste message
How it works
It reviews your text for common typos and grammar errors, such as duplicated words or subject-verb agreement issues. If requested, it also removes specific tracking parameters from links.
Installation
Easiest way: copy the skill link and paste it into Claude, ChatGPT or Cursor asking it to follow the instructions.
In the terminal (Claude Code, Codex, Cursor), in your project folder:
npx skills add sickn33/agentic-awesome-skills --skill email-issue-fixerOr download just the file:
curl -L https://raw.githubusercontent.com/sickn33/agentic-awesome-skills/main/plugins/agentic-awesome-skills-claude/skills/email-issue-fixer/SKILL.md -o SKILL.mdTechnical content (original SKILL.md)
Original text by the author, straight from GitHub.
Email Issue Fixer
Fix the small mistakes that make an email look careless, without touching the writer's voice or what the email commits to. Always return the corrected draft plus a list of what changed.
When to Use
- Use when the user asks to "fix my email", "check the grammar before I send", "proofread this email", or any pre-send cleanup of a draft.
- Use when the user also asks to "clean the links" or remove tracking from the URLs in an email. Link cleaning is never automatic.
- Do not use for casual rewrites, tone makeovers, or structural restyling that the user did not ask for.
Pass 1: correctness only
Fix these:
- Duplicated words ("a a", "the the") and wrong articles. Choose a/an by sound, not spelling: "an hour", "a university", "a one-time offer".
- Subject-verb agreement ("the reports is done" becomes "the reports are done").
- Wrong-word slips: your/you're, its/it's, than/then, affect/effect, their/there.
- Capitalization, doubled spaces, missing or doubled punctuation.
Leave these alone. They are voice, not errors:
- Sentence fragments, contractions, one-line paragraphs.
- Lowercase greetings and sign-offs, informal phrasing, slang the writer chose.
- Repetition used for emphasis.
If you cannot tell whether something is a mistake or a choice, leave it and mention it in the summary instead of changing it.
Never change names, numbers, dates, quoted text, or anything that alters what the email promises or asks for.
Pass 2: links (only when the user asks)
Do not touch links during a normal proofread. Clean them only when the user asks for it.
When asked, remove only these unambiguous tracking keys:
utm_source,utm_medium,utm_campaign,utm_term,utm_content,mc_cid,mc_eid,fbclid,gclid,igshid,_hsenc,_hsmi
Keep every other parameter. Values like id, page, v, q, or anything unfamiliar are often what makes the link work. When in doubt, keep it.
Treat ref, ref_src, and similarly generic keys as ambiguous: they are sometimes attribution, sometimes application state the link needs to work. Do not strip them automatically. Only remove them when context confirms they are attribution, or ask the writer first.
Do not delete whole links. Do not delete a link because it appears twice; signature links and repeated calls to action are usually deliberate.
Flag rather than fix: shortened links (bit.ly and similar), unfamiliar domains, and any query string that looks like a token, session ID, or credential. Point them out and let the writer decide.
If a query parameter looks like a credential, token, or session ID, do not echo its value anywhere in the output. Keep the link only by replacing the sensitive value with a placeholder such as [REDACTED] (for example ?auth=[REDACTED]), list the redaction in "Changes made", and tell the writer to re-enter the real value before sending. If they prefer, advise rotating the leaked value first.
Always return
- The corrected email, in the same format and layout it arrived in.
- A "Changes made" list, one line per change.
- A "Left alone" list for anything you judged to be voice or an ambiguous link.
- If nothing needed fixing, say so plainly rather than inventing changes.
Never
- Send, schedule, or file the email. Hand the draft back for the writer to send.
- Repeat credentials, tokens, or personal details found inside links.
- Rewrite for tone, length, or structure unless asked. If the user wants that, ask what direction they want first.
Examples
Articles and agreement
Input:
Please find attached a a copy of the report. Our sales team achieved a amazing quarter.
Output:
Please find attached a copy of the report. Our sales team achieved an amazing quarter.
Changes made:
- a a to a (duplicated word)
- a amazing to an amazing (vowel sound)
Link cleaning
Input:
https://app.example.com/reports?utm_source=newsletter&utm_medium=email&id=123
Output:
https://app.example.com/reports?id=123
Changes made:
- Removed 2 tracking parameters (utm_source, utm_medium)
- Kept id=123, which the link needs to open the right report
Limitations
This reads and edits text only. It cannot judge whether the email's facts, tone, or timing are right. That stays with the writer.
FAQ
What does the agent not do?
It does not rewrite the email to change its tone, length, or structure unless you specifically ask. It also does not send the email for you.
What are tracking codes in links?
These are pieces of information added to the end of a link (after the '?') that help track where a click came from, such as from a specific email or advertisement.
Will the agent change the meaning of my email?
No, it only corrects minor errors and preserves your voice and intended message. It does not change names, dates, or what the email promises.
Details
Original description by the author
Fix small email mistakes without touching the writer's voice, and strip tracking parameters from links on request. Always returns the corrected draft plus a change list.
Popularity
- GitHub stars
- 48k
- Ranking position
- #2908
- License
- MIT
- Updated
- 10/15/2026
Comments
- …
More skills from this repository
00-andruia-consultant
Helps your agent start or improve AI projects. It analyzes code, suggests tools, and creates action plans in Spanish.
sickn33/agentic-awesome-skills
007
Lets your agent perform security audits and threat modeling. It's useful for protecting projects against attacks and vulnerabilities.
sickn33/agentic-awesome-skills
10-andruia-skill-smith
Helps your agent create or improve AI tools. It defines how tools work, writes documentation, and validates the structure.
sickn33/agentic-awesome-skills
20-andruia-niche-intelligence
Lets your agent create a summary about a specific industry. It's useful for understanding the context, rules, usage patterns, and risks of a project, using reliable sources.
sickn33/agentic-awesome-skills
Similar skills
intent-driven-development
Helps your agent turn ideas into clear, testable plans. Defines criteria to ensure changes are made correctly and safely.
affaan-m/ECC
product-capability
Guides your agent in turning product ideas into clear plans. It defines what needs to be done, the limits, and decisions before development.
affaan-m/ECC
article-writing
Writes long-form content like articles, guides, and blog posts with a unique style. It's useful when you need well-written texts with a consistent voice and credibility.
affaan-m/ECC