open-source · MIT · for AI coding agents

Less re-explaining. More doing.

Your coding agent can move fast. It still needs a place to leave the thread. ai-skills keeps project memory close to the code and gives big jobs a way to split up.

$ /plugin marketplace add amjad1233/ai-skills

@amjad1233 is the marketplace name — not a GitHub handle.

claude-code — zsh
/plugin install continuation@amjad1233
✓ memory across sessions · ready
/continue
✓ resumed · git state verified in 5s
› spin up an agent army for these
  five changes
⚡ launching 5 agents across 5 worktrees…
Keep the thread
write it down once →
01 · Pick oneStart with the project that makes you re-explain yourself most.
02 · InstallCopy the command. Add one skill. Stop there.
03 · ResumeNext session: run /continue and get back to work.
// the problem

Claude Code is amazing — until the session ends or the work piles up.

“It forgets.”

You spent an hour getting Claude up to speed on your codebase. Tomorrow you open the project and it has no idea what you were doing — you re-explain the branch, the half-finished feature, the failing test. Ten minutes gone before you’ve written a line. And when the context window fills mid-task, you lose the thread entirely.

Fixed by continuation

“It does one thing at a time.”

You’ve got an epic split into five independent chunks, or nine changes to ship before a deadline. One Claude session grinds through them sequentially, and a single context window caps how far it gets. Subagents help — but they die when your chat does.

Fixed by agent-army
// the skills

The skills

Continuation works anywhere your agent can read a skill file. Agent Army adds Claude Code orchestration when you need parallel worktrees.

Diagram of agent-army: a master command centre coordinating parallel Claude agents, each in its own git worktree, merging into one integration branch
agent-army
Orchestration

Run a fleet of autonomous Claude agents in parallel.

The problem it kills: Big multi-stream work done one-at-a-time is slow — and one context window is the ceiling.
How it works
One agent per task — each in its own git worktree, building in parallel.
On-disk command centre survives context limits and session handoffs.
Two-signal health checks catch silent stalls and failed launches.
Intake Gate talks you out of a fleet when a cheaper tool fits.
$ /plugin install agent-army@amjad1233
Diagram of continuation: a session handoff note saved to disk on Day 1 and read back on Day 2, so Claude resumes with no recap Day 1 Day 2 · resumed
continuation
Sessions

Project memory for any coding agent — decisions, learnings, conventions and handoffs in .continuation/, written as you work, read in one hot file every session.

The problem it kills: Every new session starts from amnesia; context limits make you lose your place.
How it works
/end — save a handoff: done, pending, next step, key files.
/continue — resume in ~5s with your git state checked.
/next — open the next session in a fresh, worktree-aware terminal.
Local-only, zero dependencies — nothing leaves your machine.
$ /plugin install continuation@amjad1233

More skills incoming

A new skill is just a subdirectory and one line of config. Contributions welcome.

Star & contribute
// how it works

How a skill works

A skill teaches Claude Code a new way of working — a workflow, a set of commands — that it loads automatically, exactly when it’s relevant. You don’t memorise anything.

The ai-skills marketplace: a socket of modular skill tiles, two installed and empty slots ready for more + skill
STEP 01

Add the marketplace

One time. Registers @amjad1233 with your Claude Code.

/plugin marketplace add amjad1233/ai-skills
STEP 02

Install a skill

Whichever you want. Installs in seconds, ready immediately.

/plugin install agent-army@amjad1233
STEP 03

Just ask Claude

In plain English. The skill kicks in when it fits — nothing to memorise.

“Claude, pick up where we left off and ship the next three changes.”
// get started

Get started in ~30 seconds

Never installed a Claude skill? This is the whole thing.

01

Get Claude Code

Anthropic’s agentic coding CLI. Install and launch it if you haven’t.

$ claude
02

Add the marketplace

One time only. Registers the @amjad1233 marketplace with Claude Code.

$ /plugin marketplace add amjad1233/ai-skills
03

Install a skill

Pick whichever you want — it installs in seconds.

$ /plugin install continuation@amjad1233
04

Use it

A slash command, or just ask in plain English. The skill kicks in when it fits.

“/continue”  ·  “spin up an agent army for these five changes”
// why ai-skills

Why ai-skills

Battle-tested, not vibes.

Every skill is hardened on real multi-day work — documented lessons, not guesses.

Open-source & MIT.

Read it, fork it, ship it. No lock-in.

Zero overhead.

Skills load only when relevant and stay out of your way.

One marketplace, growing.

Add it once; new skills appear as you go.

More skills incoming

This marketplace grows. A new skill is just another subdirectory and one line of config. Star the repo to catch them, or open a PR with your own.