Skills

A collection of agent skills.

Most large skills use a thin SKILL.md entrypoint plus topic packs under references/. Load the entry file first, then only the relevant reference files for the current task.

Install

Global (all projects)

# Install all skills globally
npx skills add zzci/skills --global

# Install a specific skill globally
npx skills add zzci/skills --skill pma --global
npx skills add zzci/skills --skill pma-cr --global
npx skills add zzci/skills --skill pma-web --global
npx skills add zzci/skills --skill pma-rust --global
npx skills add zzci/skills --skill pma-go --global
npx skills add zzci/skills --skill pma-bun --global
npx skills add zzci/skills --skill pma-mem --global
npx skills add zzci/skills --skill pma-d2 --global
npx skills add zzci/skills --skill pma-draw --global
npx skills add zzci/skills --skill pma-design --global
npx skills add zzci/skills --skill bkd --global
npx skills add zzci/skills --skill gitea --global

Project (current project only)

# Install all skills to current project
npx skills add zzci/skills

# Install a specific skill to current project
npx skills add zzci/skills --skill pma
npx skills add zzci/skills --skill pma-cr
npx skills add zzci/skills --skill pma-web
npx skills add zzci/skills --skill pma-rust
npx skills add zzci/skills --skill pma-go
npx skills add zzci/skills --skill pma-bun
npx skills add zzci/skills --skill pma-mem
npx skills add zzci/skills --skill pma-d2
npx skills add zzci/skills --skill pma-draw
npx skills add zzci/skills --skill pma-design
npx skills add zzci/skills --skill bkd
npx skills add zzci/skills --skill gitea

# List available skills
npx skills add zzci/skills --list

Add PMA

First-time PMA setup in a project:

1. Ensure AGENTS.md has a ## Project Development section that references /pma and project-specific docs/paths. Use a Chinese heading only if the user explicitly requests it.
2. Create CLAUDE.md as a symlink to AGENTS.md so project instructions do not drift.
3. Ensure docs/task/index.md exists.
4. Ensure docs/plan/index.md exists.
5. Ensure core docs exist: docs/architecture.md and docs/changelog.md.

Add PMA CR

For stack-aware review after implementation and before merge:

1. Use /pma-cr for local diff review.
2. Use /pma-cr <PR-number-or-URL> for pull request review.
3. Use /pma-cr audit, /pma-cr repo, or /pma-cr --repo for repository-wide audit.
4. The skill applies shared review policy plus only the relevant stack packs:
   • TypeScript frontend
   • TypeScript backend / Bun
   • Go
   • Rust
   • Python
5. Repository audit reports findings by P0 to P3, plus coverage gaps and recommended next actions.
6. Repository audit explicitly checks isolated dead code such as orphan modules, dead handlers, stale feature-flag paths, and unreachable subsystems.
7. Dead-code output is split into Dead Code Findings, Dead Code Removal Candidates, and Needs Runtime Verification.
8. Keep findings focused on correctness, security, regressions, and operational risk.
9. Prefer repository-specific guidance from CLAUDE.md and AGENTS.md when present.

Add PMA-Web

For frontend projects using the PMA-Web stack:

1. Use /pma for investigate -> proposal -> implement workflow.
2. Default to a single React 19 + TypeScript + Vite 8 SPA; promote to a Bun monorepo only when multiple apps or shared packages exist.
3. Standardize app structure around src/app, src/features, and src/shared.
4. Use TanStack Router, TanStack Query, Zustand, Tailwind CSS v4, and shadcn/ui base-nova with @base-ui/react.
5. Configure required quality gates: lint, typecheck, build, and test.
6. Do not introduce Radix UI or other UI component ecosystems.

Add PMA-Rust

For Rust multi-crate workspace projects:

1. Use /pma for investigate -> proposal -> implement workflow.
2. Set up Cargo workspace defaults in [workspace.dependencies].
3. Pin a stable toolchain via rust-toolchain.toml and reproducible rustflags.
4. Configure strict quality gates: fmt, cranky, deny, test, and release build.
5. Choose data access strategy: SQLx (Default), SeaORM (Alternative), or diesel-async when compile-time schema typing is paramount.
6. Standardize on figment + clap, Tokio, Axum, and rustls-based networking.
7. Set up OpenTelemetry observability when deploying production services.

Add PMA-Go

For Go service and CLI projects:

1. Use /pma for investigate -> proposal -> implement workflow.
2. Use cmd/ + internal/ project layout; avoid /pkg.
3. Configure strict quality gates: goimports, golangci-lint v2, go vet, test, build, mod tidy.
4. Choose data access strategy: sqlc + pgx (Default) or GORM (Alternative).
5. Set up layered config with koanf: defaults -> file -> env -> flags.
6. Standardize on stdlib net/http + Chi (Default) or Gin (Alternative).
7. Set up OpenTelemetry observability when deploying production services.

Add PMA-Bun

For Bun backend and full-stack projects:

1. Use /pma for investigate -> proposal -> implement workflow.
2. Default to a single API project; promote to a Bun monorepo only when multiple deployable apps or shared packages exist.
3. Use backend modules under src/modules and shared code under src/shared.
4. Configure strict quality gates: lint, typecheck, build, test, coverage, and security review.
5. Choose data access strategy: Drizzle + SQLite-first storage (Default) or PostgreSQL/libSQL for multi-instance requirements.
6. Validate environment configuration with Zod at startup and centralize runtime path resolution.
7. Standardize on OpenAPIHono on top of Bun.serve(), with app.ts / index.ts / dev.ts split by runtime role.
8. Use a dedicated compile pipeline when shipping standalone binaries with embedded assets or migrations.

Add PMA-Design

For HTML design artifacts (mockups, prototypes, wireframes, decks, design systems):

1. Use /pma-design as a standalone skill — it does not require the /pma workflow.
2. Output lives under designs/<project>/ as self-contained HTML (React 18 + Babel multi-file prototypes), with React/Babel referenced locally from designs/_vendor/.
3. Clarifying questions are asked as numbered lists in chat (structured ask tools are disabled).
4. Preview is served with nsl serve --list --name <project>-designs designs inside a tmux session; deliverables are reviewed over http://<name>.localhost/..., never file://.
5. Design systems are authored or consumed via the bundled compiler/importer scripts (agents/*.mjs), bound per project through _ds/<slug>/ and _d_meta.json.
6. Harness-specific preview/screenshot/verification tools are resolved from references/claude.md (Claude Code) or references/codex.md (Codex Agent).

Add BKD

For operating BKD kanban boards via REST API:

1. Use /bkd for single issue CRUD, cron jobs, and basic operations.
2. Use short activation phrases such as use bkd to start coordination or start BKD L1 to start the three-tier L1/L2/L3 coordination pattern.
3. For long-running autonomous coordination, L1 talks to the user, L2 owns DAG decomposition and cron-driven dispatch, and L3 issues execute short-lived subtasks.
4. For multi-subtask orchestration, the skill supports two modes:
   • Worktree mode: subtasks work in isolated branches (bkd/{issueId}), suitable for multi-file changes or overlapping subtasks.
   • Simple mode: subtasks work directly on the main branch, suitable for small independent changes.
5. Pipeline-style quality assessment: each subtask is evaluated immediately on completion via logs filter, then self-review and coordinator assessment.
6. Worktree mode includes branch merge strategies with post-merge build/test verification.
7. Reference packs are loaded on demand: rest-api.md, orchestration.md, quality-review.md, merge-strategy.md, three-tier-coordination.md.

Available Skills