A growing library of Claude Code skills and specialized agents for shipping production software with an AI-assisted dev loop. One install, one ~/.claude/ directory, every skill available as a slash command.

Built for teams that treat their AI workflows as real engineering — where specs get multi-lens reviewed before code is written, and every ticket goes through a plan → build → test → review → ship pipeline.

Chain them: /council the spec → fix the issues → /build the feature.

Open Claude Code and paste this. Claude does the rest.

Install the skills repo: run git clone --depth 1 https://github.com/websentry-ai/skills.git /tmp/skills && /tmp/skills/install.sh --force && rm -rf /tmp/skills — this installs all commands (/build, /council) and all agents (principal-architect, principal-engineer, elite-pr-reviewer, code-simplifier, principal-engineer-review, product-vision-scout, ciso-evaluator, eng-leader-evaluator) into ~/.claude/. Then verify by confirming /build and /council are available as slash commands.

That's it. No manual terminal steps.

git clone https://github.com/websentry-ai/skills.git
cd skills
./install.sh

The installer globs agents/*.md and commands/*.md — any skill added to this repo will be installed automatically.

Want teammates to get the skills automatically when they open Claude Code in your project?

# From your project root
mkdir -p .claude/commands .claude/agents
cp ~/.claude/commands/build.md .claude/commands/
cp ~/.claude/commands/council.md .claude/commands/
cp ~/.claude/agents/*.md .claude/agents/
git add .claude/ && git commit -m "Vendor skills for AI-assisted development"

Now every teammate who opens Claude Code in this repo gets /build and /council with zero setup.

/council PLAN.md
/council https://linear.app/unbound/issue/ENG-1234
/council <paste spec inline>

Convenes five reviewers in parallel on a spec or plan, then fuses the critiques into one verdict.

Output:

• Verdict — BUILD / BUILD-WITH-CHANGES / RETHINK / KILL
• Consensus items — auto-escalated when two or more lenses flag the same issue
• Dissents — surfaced honestly, not papered over
• Sharpened Definition of Done — verifiable bullets you can test against
• Scenarios to run — concrete (often non-deterministic) test scenarios
• Open questions — what the author must answer before /build

• Before every /build on a non-trivial feature
• When a spec feels thin and you want forcing questions
• When the team disagrees on scope and you need the disagreement surfaced
• Before a design doc goes to review — catch issues while the spec is cheap to change

• All five lenses run in parallel — one message, five subagents, one synthesis
• Consensus auto-escalates — two lenses raising the same concern = blocker, not nice-to-have
• Dissents are the point — when the lenses disagree, that's high-signal; /council surfaces the disagreement instead of averaging it out
• No code is written — /council reviews specs, /build writes code

/build Add a REST endpoint for user preferences with CRUD operations

A production-grade CI pipeline running inside your Claude Code conversation. One command. Plan → Code → Simplify → Test → Review → Fix → Ship → Review Again → Validate.

• Build mode — no code exists yet. Starts from planning (Step 0).
• Iteration mode — code already exists (dirty tree or commits ahead). Skips to simplification (Step 2).

• Never skip tests — every code-changing step re-runs the suite
• Never skip reviews — both review passes are mandatory
• Tests at the outermost layer — API/task level, not helper unit tests
• User approval required on the plan before coding
• Fail-stop — if any step fails, stop and report

Every agent is also usable standalone via the Agent tool or direct invocation.

• Claude Code CLI installed and authenticated
• GitHub CLI (gh) — used by /build for PR creation
• A git repository with a remote (for /build)

• gstack — /build auto-detects /ship and uses it for PR creation with version bump + CHANGELOG + bisectable commits.
• Playwright MCP — product-vision-scout uses it to navigate and screenshot live product.
• Linear MCP — /council can consume Linear ticket URLs directly.

The repo is designed to grow. To add a new skill:

1. Drop the command file in commands/{name}.md (with frontmatter description:).
2. Drop any new agents in agents/{name}.md (with frontmatter name:, description:).
3. Add a row to the Skills table at the top of this README.
4. Add a section below describing the command.

install.sh auto-picks up everything in commands/*.md and agents/*.md — no script changes needed.

Edit the agent .md files in ~/.claude/agents/ to change code style standards, testing philosophy, or add framework-specific guidelines.

Edit ~/.claude/commands/build.md or council.md to skip steps, add steps, or change hard gates.

Add project-specific instructions in your project's CLAUDE.md. Agents respect CLAUDE.md conventions, so you can override behavior per-repo without modifying global agent definitions.

Each command is a Claude Code slash command that orchestrates one or more agents via the Agent tool. Agents run as subagents with their own context and specialized prompts.

/build uses hard quality gates — if tests fail after Step 3 or Step 5, the pipeline stops instead of pushing through.

/council runs all five lenses in parallel in a single message, then synthesizes. No lens sees another lens's output — critiques stay independent.

MIT