Claude Skills by imoonkey

Align the design of the system or anything else between Codex and Claude.

Systematic code review with severity-based findings. Use before merging or when reviewing changes.

Coding conventions and patterns. Auto-applied as reference during code changes. Detects the project stack and loads stack-specific idioms.

System design from first principles. Produce a design doc, self-review, iterate. Use before `/eng-plan-review` or `/implement`, for non-trivial work.

Reply in place to `>`-marked comments inside a markdown doc, keeping multi-round Q&A navigable. Use when the user runs `/discuss` or left notes/questions in a doc to answer inline rather than in chat.

Two agents (Claude + Codex) independently design, cross-review, then align. Use for critical design decisions that benefit from diverse perspectives.

Review an implementation plan before coding — for work that touches multiple files, changes data flow or state, or adds integrations. Use after `/design` or on any rough plan.

Write an implementation summary for landed work — top-down outcome and structure, never a chronological commit log. Use when a milestone lands or when asked to summarize implemented work.

Full implementation workflow — plan, phased build with review loops, verify. Use for any non-trivial feature or system design.

Set up a project for all AI agents (Claude, Codex, Cursor, Gemini) — CLAUDE.md, multi-tool symlinks, and a doc/ memory base. Use when onboarding a repo or the user says "init all" or "set up for codex".

Systematic debugging with root cause investigation. Use when debugging, investigating bugs, or tracing errors. Use before fixing any non-trivial bug.

Execute active tasks from the project task graph using yaco agent workers. Use when the user wants to run, advance, or check on task execution.

E2E and integration QA — derive the user flows a change affects and verify them (Playwright, HTTP, CLI) in a fix-verify loop. Use after implementation; unit tests → /tdd, /verify.

Subtractive-KISS discipline — write code pre-collapsed so no cleanup pass is needed. Use when implementing or designing code, BEFORE adding any abstraction (class, registry, config knob, mode), when reviewing your own diff, or planning a subtraction pass.

Test-driven development. Write tests FIRST, then implement. Use for core logic, state machines, data transformations.

Invoke before high-stakes or "impossible"-seeming work to push past first-pass thinking — reject the first solution that merely works, attack the problem from first principles, and find the inevitable design.

Sync docs with code changes and maintain changelog. Use after architecture/workflow changes, before or after commits.

Run verification loop (build, lint, test, security) before commits. Use before any commit or PR, or after significant changes. Detects project stack automatically.

Orchestrate the YACO agent lifecycle (Claude Code, Codex) through the yaco agent CLI. Start, send, wait, list, inspect, and link agent sessions.

Path and handoff conventions for a YACO project (companion to `yaco paths`). Auto-apply as a reference whenever a skill writes a doc, task graph, or archive in a registered YACO project.

Create and manage the project task graph via the `yaco task` CLI — plan milestones, break work into tasks, reorganize, update progress, or absorb subtasks from /design.

Git worktree lifecycle for task-isolated execution via the `yaco worktree` CLI — create/reuse, merge up the worktree/branch DAG, clean up. Use when a task carries a `worktree` slug or /orchestrate needs an isolated checkout.