Choose an Engineer Skill

Engineer Skills are outcome-led workflows for work such as planning a change, implementing it, investigating a failure, or preparing a delivery. Start with the outcome you need, then read that Skill's guide before invoking it. The guide defines its inputs, review gates, runtime notes, and authority boundaries.

Invoke a Skill

Claude Code and Cursor use the slash form; Codex uses the dollar-prefixed form for Skills that its adapter projects. For example:

/ak:brainstorm "Compare safe rollout approaches for this API change"

$ak:brainstorm "Compare safe rollout approaches for this API change"

Replace the Skill name and request with the workflow you choose. Do not assume that every runtime projects the same Agents, Hooks, or provider tools; check the individual guide and Runtime adapters.

Codex intentionally excludes ak:team because it does not provide the complete Agent Teams lifecycle. A Codex install reports this under Capabilities excluded (unsupported on this target) and does not create a $ak:team invocation or substitute ordinary subagents.

Start here

Chain Skills into a workflow

Each Skill invocation is a separate, reviewable workflow. Carry the prior Skill's decisions, evidence, artifacts, and unresolved questions into the next request. Do not treat a suggested chain as authority to skip a gate, publish, deploy, spend, or make another external change. Use ak:handoff when the chain continues in another session or Kit.

Continue with a detailed guide:

• Feature delivery from discovery through shipping
• Evidence-led bug diagnosis and repair
• UI design, implementation, browser verification, and review
• Cross-Kit product-launch handoff

Stop the chain when an input is missing, a review rejects the result, or the next step needs authority you do not have. Resume from the failed prerequisite; do not silently continue with assumptions.

Choose by outcome

Frame and plan work

• ak:ask answers a focused technical question from current project evidence without starting implementation.
• ak:brainstorm compares viable approaches and produces a bounded direction.
• ak:research gathers current primary evidence for a technical question and preserves uncertainty.
• ak:plan produces a phased implementation roadmap with risks, dependencies, and completion checks.
• ak:issue-to-plan audits a GitHub issue against the repository and stops at its decision gate before implementation.

Build systems

• ak:cook implements a clear task or approved plan with testing and review in the workflow.
• ak:frontend-development handles scoped React and TypeScript interface work using the project's existing conventions.
• ak:backend-development designs and verifies a bounded backend, API, or data-access change.
• ak:databases covers data structures, query behavior, and guarded migration or administration preparation.
• ak:devops prepares and verifies infrastructure, CI/CD, or GitOps work behind deployment and deletion gates.

Diagnose and validate

• ak:debug builds an evidence chain to a root cause.
• ak:fix implements the smallest repair aligned with a proven cause.
• ak:test runs code or UI validation and reports failures without starting fixes.
• ak:code-review reviews a diff, pull request, commit, workspace, or codebase for production risk.
• ak:security-scan performs a read-only scan for secrets, dependency issues, and code patterns, then verifies likely matches.

Design and explain

• ak:frontend-design turns a brief or visual reference into an implemented interface with visual and accessibility gates.
• ak:ui-ux-pro-max develops bundled UI recommendations and persists approved design rules.
• ak:docs creates, reconciles, or summarizes the smallest useful project documentation route.
• ak:mermaidjs-v11 authors and verifies Mermaid v11 diagrams for systems, processes, data, timelines, or journeys.
• ak:preview opens existing content or creates visual explanations, reviews, diagrams, slides, and recaps.

Deliver and preserve work

• ak:git stages, commits, pushes, opens pull requests, or merges through explicit Git safety gates.
• ak:deploy selects a supported host, deploys with explicit provider authority, verifies the URL, and records rollback guidance.
• ak:ship runs release gates and prepares a completed branch for a pull request.
• ak:handoff records goals, decisions, evidence, blockers, and open state for a fresh session.
• ak:worktree creates and audits isolated Git worktrees, with cleanup separated behind review.

Extend the toolchain

• ak:use-mcp discovers MCP tools, inspects their schemas, and keeps external effects controlled.
• ak:mcp-builder researches an API, designs agent-centered tools, implements an MCP server, and verifies it.
• ak:skill-creator scopes, scaffolds, validates, evaluates, and packages a reusable Skill.
• ak:find-skills routes to an installed Skill or searches for a reviewed external package with approval before installation.
• ak:agent-browser drives a local or cloud browser with explicit session boundaries and reviewable artifacts.

Complete public catalog (97 Skills)

The table below mirrors the reviewed Engineer public catalog. Every entry opens the corresponding Skill guide; source aliases and internal-only identities are not presented as separate public workflows. The catalog describes the Kit source, while the projected set can be smaller for a runtime; in particular, Codex omits ak:team.

Keep the Kit boundary visible

Some names, including ak:ask, ak:plan, ak:cook, and ak:preview, also appear in Marketing Kit. A shared name does not mean the two Kits ship the same workflow. Use the guide under the Kit you installed and keep that Kit's inputs, gates, and runtime notes attached to the task.

Skills are only one layer of Engineer Kit. Review Engineer Agents for specialist roles, Hooks and runtime for projected automation, and Lifecycle and recovery for refresh, uninstall, and ownership behavior. Use Installing kits to choose target, scope, delivery mode, and installed Skill selection, or return to the Engineer Kit overview.