Open source · Claude Code
claude-skills-pm
Seven Claude Code skills for the work a PM actually does.
Outreach follow-ups, LinkedIn drafts, contract review, dispute letters, interview prep, long-form articles, and a writing audit that catches the machine tells before anything goes out. Every one of them started as a skill I use daily. They have been de-identified so you can anchor them in your work instead of inheriting mine.
git clone https://github.com/debashisn94/claude-skills-pm.git MIT licensed. v0.1, seven skills, stable in daily use.
A skill is a folder. The value is in what it says.
A Claude Code skill is a folder with a SKILL.md in it. Claude reads the description, works out when the skill applies, and follows the instructions inside. That is the whole mechanism, and it is why the interesting part is the writing rather than the plumbing.
Each of these came out of my own week and got edited every time it produced something I would not have sent. Hundreds of real runs across job applications, vendor disputes, LinkedIn posts, Medium drafts, cold emails and contract reviews. What survived that is what sits in the repo.
Then I stripped my data out of them. The examples carry placeholders like [your metric] and [counterparty] on purpose, so the structure travels and the specifics stay yours.
The seven.
Invoke one directly with a slash command, or describe the task and let Claude route to it.
anti-ai-audit A 14-section audit against the patterns that make writing read as machine-written: banned words, em dashes, neat tricolons, callback closers, endings that tie up too tidily. It runs before anything public-facing ships. The full style guide travels with it as a reference file the audit loads.
linkedin-post LinkedIn drafts in your voice. Tone-mode aware, so a career post and a craft post do not come out sounding identical, and it hands the result to anti-ai-audit before you ever see it.
cold-outreach-followup The D+3, D+10 and D+21 follow-up sequence for job applications and prospecting. Each touch escalates the ask instead of repeating the last email in different words.
dispute-letter Dispute letters written to survive a tribunal. Evidence cited, a number anchored, a deadline set, escalation signalled. Deposits, refunds, billing, vendors.
contract-review Risk flagging for MSAs, NDAs, employment agreements, contractor ICAs and severance. It marks the clauses worth arguing over and drafts the counter-ask.
pm-prep PM interview prep: case studies, product design, A/B test design, KPI trees, RICE, market segmentation, behavioural stories.
medium-article Long-form drafts in your voice, taken to publication-ready. Also handles the outline-to-article expansion when you already know the shape.
Two commands and a restart.
git clone https://github.com/debashisn94/claude-skills-pm.git Anywhere you like. The repo is small, seven folders and a licence.
cp -r claude-skills-pm/skills/* ~/.claude/skills/ Drops each skill folder into your own skills directory.
Restart Claude Code and the seven show up. Call one with /skill-name, or just say what you need and let it pick.
Prefer to read them before you run them? Every SKILL.md is plain markdown on GitHub.
How the repo is laid out.
Each skill is a folder with a SKILL.md plus whatever reference docs it needs. Nothing gets compiled, so an edit takes effect the next time Claude Code loads.
skills/
anti-ai-audit/
SKILL.md
references/anti-ai-writing-style.md
cold-outreach-followup/SKILL.md
contract-review/SKILL.md
dispute-letter/SKILL.md
linkedin-post/SKILL.md
medium-article/SKILL.md
pm-prep/SKILL.md
voice-corpus/
README.mdMake them yours.
These carry structure, not my personal data. A dispute letter skill that already knows my landlord is useless to you. So the examples are blanks, and the first thing worth doing is filling them with your own stories, numbers and phrasing.
The three voice-anchored skills, anti-ai-audit, linkedin-post and medium-article, get noticeably sharper if you add a voice corpus: a small folder of writing you published before you started using AI. It is optional and everything works without it. I think the corpus is where most of the gain sits, though it is honestly the part I have tested least on writing other than my own.
The convention for it lives in voice-corpus/README.md.
Where it stands.
All seven are in daily use and de-identified for public use. MIT licensed, so fork and republish as you like.
Drafts scored against your own published writing rather than checked against a list of tells. No date on it yet.
Install them, then start cutting the bits that do not sound like you.
If a skill gets something wrong for your work, that is a signal about your work, and the fix is usually four lines in a markdown file. Open an issue if you want a second pair of eyes on it.
cp -r claude-skills-pm/skills/* ~/.claude/skills/