Loop AI Skills

Six free, safe, single-purpose skills for Claude Code. One per video on the Loop AI channel. No telemetry, no network calls, nothing destructive: every skill is read-only except one clearly named file, and asks before writing anywhere else.

Install (one command)

Option A: copy the skills (works everywhere)

git clone https://github.com/DimitriSmagliy/loop-ai-skills && mkdir -p ~/.claude/skills && cp -r loop-ai-skills/plugin/skills/* ~/.claude/skills/

Use .claude/skills/ inside a repo instead of ~/.claude/skills/ to install for one project only. Skills are then called by plain name, e.g. /handoff.

Option B: as a plugin from a marketplace

/plugin marketplace add DimitriSmagliy/loop-ai-skills
/plugin install loop-ai-skills@loop-ai

Skills are then namespaced, e.g. /loop-ai-skills:handoff. Try without installing: claude --plugin-dir ./plugin.

The skills

Skill	What it does	Writes
/handoff	Writes or updates HANDOFF.md so a fresh session continues where you stopped	HANDOFF.md
/secret-guard	Scans for committed secrets and checks .claude/settings.local.json is gitignored	nothing (asks before touching .gitignore)
/ship-check	Pre-commit checklist: tests, lint, diff review, secrets, debug logs. Verdict: SHIP or FIX FIRST	nothing
/explain-like-new	Explains an unfamiliar repo and builds CODEMAP.md (topic to file:line)	CODEMAP.md
/deny-rules	Merges sensible deny rules (.env, rm -rf, force push, prod hosts) into .claude/settings.json, explaining each	.claude/settings.json after you confirm
/cost-saver	Reads your /context and /usage, trims context, tells you when to /clear	nothing

Why secret-guard matters: .claude/settings.local.json stores your "don't ask again" approvals, and a committed copy can leak credentials that were pasted into commands.

Safety model

• Each skill lists its allowed-tools, kept to reads and read-only git commands.
• Nothing runs git commit, git push, rm -rf, history rewrites or installs.
• Secrets are never printed in full; the scanner reports file:line only.
• /deny-rules only runs when you type it (disable-model-invocation).
• Read every SKILL.md before installing. They are short on purpose.

Compatibility

Built against the official Claude Code skills and plugin docs (code.claude.com, fetched 2026-11-10) and tested on Claude Code 2.1.322. The skills use the Agent Skills SKILL.md format.

License

MIT, see LICENSE.

Links

Landing page: https://dimitrismagliy.github.io/loop-ai-skills/

Not affiliated with or endorsed by Anthropic.