PM is a free skill for Claude Code and GitHub Copilot that adds a project manager to your workflow. It intercepts every prompt, builds a structured spec before writing any code, and only touches the files in that spec — nothing else. It runs security scans after every build, auto-saves your progress, and maintains a single source of truth for your design tokens. Say pm on to activate.

Claude Code: Run mkdir -p ~/.claude/skills/pm && curl -o ~/.claude/skills/pm/SKILL.md https://raw.githubusercontent.com/christinezhang5/project-manager-nuggieai-skill/main/skills/pm/SKILL.md in your terminal, then add your token to CLAUDE.md.

Copilot: Add your token to .github/copilot-instructions.md, select Claude as your model in agent mode, and say pm on. Create a free account to get your token.

Yes. The 4 core features (pm prompt, pm global, pm scan, pm revert) plus 5 modules (pm standup, pm scope, pm map, pm style, pm merge) are free forever. You also get a 15-day free trial of the 3 pro modules (pm guard, pm connect, pm claudemd). PM Pro at $0.99/month keeps them permanently.

It intercepts every request before code runs and builds a structured build spec. It scans your files to estimate confidence (0-100%), reads specific files if unsure, and asks at most one question. You see exactly which files will change and what will be left alone before anything happens. This prevents Claude from touching the wrong files or over-engineering a simple fix.

Those are the 5 build modes pm prompt assigns to every spec. They show up in your dashboard activity log so you can see what kind of work each session was made of. PM always picks the smallest mode that fits — fewest files, fewest changes, least abstraction.

surgical — a tiny, targeted change to one thing. Fix a bug, change a value, rename a variable. Triggered by words like fix, change, update.

feature — building something new. A new button, page, endpoint, or behavior that didn't exist before.

style — visual or CSS changes only. Colors, spacing, fonts, layout. No logic changes. Auto-loads pm style when a CSS file or inline style is touched.

refactor — restructuring existing code without changing what it does. Splitting a long file, extracting a helper, cleaning up duplication.

redesign — large rework. Rebuilding or replacing a whole chunk of the app. Bigger blast radius than refactor.

pm guard runs a deep 18-point security audit after every build. It checks for exposed API keys, SQL injection, XSS, plain text passwords, open routes, and more. Seven of the most dangerous issues get auto-fixed silently. The rest get flagged with a plain English question like "Your signup page has no rate limit — want me to add one?" Score starts at 100 and deducts per failing check.

Yes. pm revert does three things Claude can't do on its own: it auto-saves every 5 prompts so you never lose work, it lets you search by description (pm go back to when the login worked) instead of counting steps, and it lets you name checkpoints (save as auth-working) so you can bookmark known-good states. No git required.

After 15 days, the 3 Pro modules (pm guard, pm connect, pm claudemd) deactivate. The 4 core features and 5 free modules (pm standup, pm scope, pm map, pm style, pm merge) stay active forever. Upgrade to Pro for $0.99/month to keep everything.

Yes. PM works in GitHub Copilot's agent mode with Claude as the model. Add your token to .github/copilot-instructions.md instead of CLAUDE.md, select Claude in the model picker, switch to agent mode, and say pm on. All commands work the same way.

pm global maintains a globals.json file as the single source of truth for every design token in your project — colors, fonts, spacing, API URLs, env values. When it finds a hardcoded value like a hex color, it silently replaces it with a reference to globals.json. When the same value appears in 2+ files, it promotes it to a shared token. This prevents the "I changed the brand color but missed 12 other places" problem.

No. pm revert handles version control with plain English commands. pm branch creates folder-based copies of your project for risky experiments. Everything is automatic — no commits, no branches, no merge conflicts.

Your skill token (starts with nvsk_free_) is a personal key that authenticates your PM session. It lives in your project's CLAUDE.md or copilot-instructions.md file. The token only grants access to PM — it cannot read your code, access your files, or do anything else. Same token works in unlimited projects.

Go to your dashboard → account settings → copy your token fresh using the copy button. Paste the full token (including the nvsk_free_ prefix) into your CLAUDE.md, replacing the old one. Make sure there's a space after the colon: skill_token: nvsk_free_xxxx

Say pm on then tell PM what you want to build in one sentence. Be specific — "add a signup form that saves to the database" works better than "make the signup work." PM will scan your files, build a spec showing exactly what it plans to change, and ask you to confirm before touching any code.

Give it one task at a time. If you have multiple things to do, PM will split them into a queue automatically. Don't over-explain — one sentence is enough. The confidence loop will read your files and ask a question if it needs more info. If the spec looks wrong, say "no, not that file" or "leave the header alone" and it'll rebuild. The spec is your chance to catch mistakes before any code runs.

pm style auto-loads whenever you edit a CSS file, but you can also say pm style on anytime. It's most useful when your styles "don't look right" and you can't figure out why — usually a stronger CSS selector is silently overriding your rule, or you have a class name typo. It catches these invisible bugs that have no error message.

Vercel's Hobby tier has a hard limit of 12 serverless function files. If you go over, your deploy fails with a confusing error. pm merge watches your api/ folder automatically — when you hit 12 files, it groups related endpoints into single files (all auth routes together, all user routes together), updates every fetch call in your frontend, and keeps you under the limit. You never have to think about it.

Say pm on then setup. PM asks 3 questions — what you're building, what tools you're using, and what to build first. It saves the answers to CLAUDE.md and immediately starts building your first task. From there, just describe what you want one thing at a time. PM handles the rest — specs, security scanning, auto-saving, globals management.

Just say "my data isn't loading" or "the form isn't submitting" — pm connect auto-loads and runs 5 checks: is the URL correct, is the request shape right, is the response shape right, are auth headers attached, and are you hitting the right server. It shows a connection map with pass/fail for every endpoint so you can see exactly which wire is broken. Pro module — included in the 15-day trial.

CLAUDE.md grows over time as PM adds project context. After many sessions it can get bloated with outdated info — "currently building the login page" from 3 weeks ago, duplicate descriptions, verbose notes. pm claudemd auto-loads when your CLAUDE.md hits 60+ lines and trims the PM-generated content while keeping everything you wrote. Less bloat means faster, cheaper sessions. Pro module — included in the 15-day trial.

PM auto-saves every 5 prompts. You can also say save or save as login-working anytime. To undo: say pm go back for the last change, pm go back 3 for 3 steps, or pm go back to when the login worked to search by description. Say pm history to see all saves.

pm standup generates a plain English session summary — what you built, what still needs fixing, and what to do first next session. It auto-triggers when you say "thanks", "bye", "done", or "pm off". It also includes a shareable paragraph you can paste into Slack or a project update. If pm scope parked any ideas during the session, they show up here too. Free forever.

At session start, PM scans your project and gives every file a short nickname — like "auth" for AuthenticationService.jsx or "dash" for DashboardPage.tsx. You can use the nickname in any prompt ("fix auth") and PM knows which file you mean. Say pm map show to see all nicknames, or pm map set login to auth to override one. Free forever.

Yes — say pm try experiment to create a full copy of your project. Work on it freely. If it works, say pm keep to merge changes back with automatic conflict checking. If it fails, say pm toss to discard it. Your original project stays untouched the whole time. No git knowledge needed.

Just say "done", "thanks", "bye", or pm off. PM auto-generates a session summary showing what you built, what needs fixing, and what to start with next time. This summary includes a shareable paragraph and any parked ideas from scope. Your CHANGELOG.txt is updated and all session state is cleared.

My Dashboard is your personal home: your skill token, the invite-a-friend referral link, your access (PM Free or PM Pro), your streak, and a "built today" share card on the left side panel. The main panel has your usage (toggle between today / this week / all time), a 14-day daily activity chart, projects you've worked on, build mode and security breakdowns, top time-savers, full session history, and your messages inbox. The side panel auto-hides on narrower screens — a small toggle to its left brings it back when you want it.

There's a pill toggle at the top of My Dashboard with two views.

Overview is the full layout — token, referral, access, streak, share card, usage, daily activity chart, projects, build modes, security, top time-savers, session history, and messages inbox. Use this when you want everything in one place.

Skill Stats is the same layout as the public Skills > Results tab but populated with your own numbers — four big savings cards (time saved, prompts saved, tokens saved, bugs caught), a 7-day activity graph, and your recent sessions. Use this for a focused, glanceable view of how PM has saved you time this week.

The Team Dashboard is a separate page (Pro only) that aggregates everyone in your team. You'll see a weekly team summary card (shareable to Slack or X), per-member leaderboard for time saved, prompts, and bugs caught, a 14-day team activity chart, the breakdown of who's using which build modes, the team's security trend, and a project-board view for tracking shared work. It's the same idea as My Dashboard but rolled up across all members of your team.

Teams are a Pro feature. Once you're on Pro, head to the Team tab in the top nav. To create: enter a team name and click "create team" — you become the team owner and get an invite code. To join: paste the invite code from someone on the team and click "join team". Owners can invite people directly by username/email from the team page or share the team URL (nuggieai.com/?team=<code>). New members start contributing to the team's stats from their next pm session.

A nuggie is a personalised PM persona — your tone, pace, code-style preferences, vocabulary, allowed Bash patterns, and project-specific triggers all bundled into one config that gets loaded alongside the core skill at session start.

To train: in Claude Code with PM active, say pm train. PM asks whether you want to train explicitly (you describe what to capture in your own words) or inexplicitly (it scans the current session — corrections, repeated patterns, vocabulary you used — and drafts sections silently). Either way the result is compressed and sent to nuggieai.com, merged into your nuggie's skill_md by section header. Use pm train change nuggie to switch active nuggie first, or pm train show to print what's currently saved.

Size limits on the saved file: 1 KB on Free, 8 KB on Pro. You can also edit it as plain markdown directly from the nuggie page on nuggieai.com.

Yes. Open their nuggie from their profile or the feed and click "copy". Pick a name for your version — your copy is saved as <original-name>/<your-name> so attribution is preserved. Once copied, the nuggie belongs to you and you can train it (training someone else's nuggie isn't allowed — you can only train ones you own). Original nuggies stay untouched. You can also just use someone else's public nuggie without copying — the "use this nuggie" button switches your active nuggie to theirs.

Your profile shows your username, display name, follower / following / users counts, all your nuggies (with use / skill / retrain / delete buttons per nuggie), a privacy toggle to make your profile public or private, and two tabs: "posts" (your shared posts to the feed) and "used" (history of nuggies you've used). Other people see your public profile when they visit nuggieai.com/<your-username> — if you're set to private, only your followers see your nuggie configs.

The articles page is a curated feed of updates and insights from the nuggieai team. Find it in the top nav when you're not signed in, or under the account dropdown when you are. Articles cover things like new features, tips for getting the most out of PM, and what we're building next.