Claude Code skills

Skills are small bundles (markdown + optional scripts) that extend Claude Code’s tool surface. Alfred-OS doesn’t ship skills itself; consumer agents pick what they need.

Full guide at docs/SKILLS.md. Highlights:

Where they live

~/.claude/skills/
├── code-review/SKILL.md
├── code-review-and-quality/SKILL.md
├── debugging-and-error-recovery/SKILL.md
├── frontend-ui-engineering/SKILL.md
├── security-and-hardening/SKILL.md
├── spec-driven-development/SKILL.md
├── autofix/SKILL.md
└── gstack/                  # gstack tap installs as a directory of subskills
    ├── browse/
    ├── investigate/
    ├── qa/
    ├── review/
    └── ship/

Recommended set for an autonomous engineering fleet

Skill	Source	Used by	Why
spec-driven-development	Anthropic	feature-dev	Forces the model to derive code from a written spec
code-review-and-quality	Anthropic	feature-dev (self-check), reviewer	Multi-axis review
security-and-hardening	Anthropic	feature-dev (auth), reviewer	Security-specific lens
debugging-and-error-recovery	Anthropic	bug-triage, monitoring	Systematic root-cause path
frontend-ui-engineering	Anthropic	feature-dev (frontend)	Component patterns
code-review	CodeRabbit	reviewer	Backbone for structured review
autofix	CodeRabbit	review-fix	Apply CodeRabbit P0/P1 fixes with per-change approval
/review, /ship, /qa, /browse, /investigate	gstack	various	gstack’s CLI-first review/ship/QA flow

Install

# Anthropic official
git clone --depth 1 https://github.com/anthropics/claude-code.git /tmp/cc
cp -R /tmp/cc/skills/* ~/.claude/skills/
rm -rf /tmp/cc

# gstack
git clone https://github.com/garrytan/gstack.git ~/.claude/skills/gstack
(cd ~/.claude/skills/gstack && ./setup)

# CodeRabbit
npx -y skills add coderabbitai/skills --global --yes \
    --agent claude-code --skill '*'

For a single fresh-install script, see docs/SKILLS.md#skill-install-automation.

Security note

Skills run with the same permissions as claude. They can read/write files in the agent’s worktree, run shell commands, invoke tools. Treat any new skill the way you’d treat any other dependency:

1. Read the SKILL.md.
2. Skim the scripts the skill might invoke.
3. Run a Snyk / CodeQL scan on unfamiliar sources.
4. Pin to a specific commit when installing from a third-party tap.

The fleet’s IAM-per-agent + per-firing-worktree-isolation patterns limit blast radius (a malicious skill in the Lucius worktree can’t reach the operator’s home or the secondary Claude account). Mitigations, not prevention.

Anti-recommendations

• Anything that auto-publishes (auto-tweet, auto-deploy, auto-merge). Use as draft-then-review only.
• Skills that fork to the network without explicit allowlists. Network egress from a worktree is a known agent attack vector.
• Skills the operator hasn’t read. Skills are markdown. Read them.

Where skills live in the framework’s mental model

Skills are operator-installed, not framework-bundled. Alfred-OS ships zero skills by default. Consumer fleets pick. Keeps the framework pluralist (different fleets, different stacks) and small (no skill maintenance burden on us).