Agent skills

Portable agent skills I use across projects, published as source.

Each skill is plain Markdown plus the scripts it calls, so it works in Claude Code, Codex, Cursor, and anything else that reads SKILL.md. Nothing here depends on one private project or one machine.

Install

With the skills CLI:

npx skills add nick-neely/agent-skills --list          # see what's here
npx skills add nick-neely/agent-skills --skill visual-verification --global

Or install by hand: clone the repo and symlink the skill directory into your agent's skills directory (~/.claude/skills/, ~/.codex/skills/, and so on).

git clone https://github.com/nick-neely/agent-skills.git
ln -s "$PWD/agent-skills/skills/visual-verification" ~/.claude/skills/

Skills that shell out have their own runtime dependencies, listed in each SKILL.md. ui-evidence checks its own up front with scripts/preflight.mjs.

Skills

• generated-image-assets - Remove an image's background, verify the alpha edges, and size it for shipping.
• implement-stack - Implement a spec, issue set, or free-form request as a reviewable stack of pull requests.
• implement-program - Deliver dependency-linked tickets through child pull requests into one human-gated integration pull request.
• ship-spec - Claude Code only. Deliver every issue in a GitHub spec as a resumable orchestrated loop.
• to-spec - Synthesize the current conversation into a spec and publish it to the issue tracker.
• to-tickets - Break a spec or plan into tracer-bullet tickets that declare their blocking edges.
• ui-evidence - Capture, prepare, annotate, and publish UI screenshots and GIFs into a pull request or issue.
• visual-verification - Screenshot, inspect, and validate local app UI, including authenticated pages.

to-spec and to-tickets are customizations of Matt Pocock's skills. See THIRD_PARTY_NOTICES.md for provenance.

Layout

skills/<skill-name>/
  SKILL.md            required
  agents/openai.yaml  required
  scripts/            optional
  references/         optional
  assets/             optional

Everything a skill references lives inside its own directory. No shared libraries between skills, no links that escape the skill root.

Validate

./scripts/validate

One gate, no dependencies beyond Node and git. It checks layout, frontmatter schema, directory-name agreement, OpenAI metadata, local links, script references, writing conventions, portability, licenses and provenance, executable modes, JavaScript syntax, and symlink containment, then runs the negative fixtures and behavior tests. CI runs it on x86_64 Linux, Apple Silicon macOS, and Windows.

Contributing

Skills are curated, not accepted by default - see CONTRIBUTING.md for what belongs here and what does not.

License

MIT, except where a skill directory carries its own LICENSE. See LICENSE and THIRD_PARTY_NOTICES.md.