agent-skills

Personal Agent Skills for Cursor and Claude Code. Install with the Skills CLI.

Install

Globally (all your projects)

# All skills, Cursor + Claude Code
npx skills add piscis/agent-skills -g -a cursor -a claude-code -y

# Or install everything to all agents
npx skills add piscis/agent-skills --all -g

# Update later
npx skills update -g

A single skill, into one project

Run this from the project root. Omitting -g installs at project scope:

npx skills add piscis/agent-skills --skill conventional-commit -a claude-code -y

That writes the skill into the project and records it in skills-lock.json:

.claude/skills/conventional-commit/SKILL.md
skills-lock.json

Commit both, and anyone else on the project gets the same skill. Update it later with npx skills update -p.

For more than one skill, repeat --skill — a comma-separated list is not accepted:

npx skills add piscis/agent-skills --skill ask-me --skill ai-review -a claude-code -y

Targeting several agents at once (-a claude-code -a cursor) installs into .agents/skills/<name>/ and symlinks each agent's directory at it, rather than copying into .claude/skills/. Skills with a references/ directory bring it along either way.

Preview first

npx skills add piscis/agent-skills --list

Skill index

Skill	Description
ai-review	Review the current branch PR and post inline findings, after confirmation (GitHub gh or Forgejo MCP).
ask-me	Surface only the open questions the user must answer to improve the current plan or task.
commit-push-pr	Create a branch if needed, commit, then push and open a PR after confirmation (GitHub gh or Forgejo MCP).
conventional-commit	Preview a Conventional Commits message, then commit only after explicit confirmation.
orpc-api	oRPC + Zod 4 procedure conventions for piscis/fritzbox-cloudflare-dyndns-vercel (schemas, typed error mapping, /api catch-all). Auto-applied.
update-issue	Refine an issue title, body, and acceptance criteria, then update after approval (GitHub or Forgejo).

Invoke explicitly (slash skill / $skill) — these use disable-model-invocation: true, except orpc-api, which the agent applies from context.

Authoring

See AGENTS.md for how to add or change skills.