---
name: better-coding-help
description: "Explain the better-coding skill set: what a good coding workflow looks like, which of the ten skills fits which task, how they chain together, and how to invoke them manually in Claude Code, Codex, Cursor, or Windsurf. Use when the user asks how these coding skills work, which skill to use for a task, how to call, install, or update them, or wants an overview of the recommended workflow. Informational only — it explains and hands off; to set the workflow up for the agent use /better-coding-start, and for actual coding work start with /better-coding-orient. Trigger: /better-coding-help"
metadata:
  version: "2.12.0"
  license: Apache-2.0
---

# Better Coding Help

The map of the skill set. This skill explains; it never plans, codes, reviews,
or verifies. After answering, hand off to the right skill for real work.

## Activation

Begin the first response after loading with exactly this line:
▸ better-coding-help active — the map of the skill set

Emit it once, then answer the user's question using the material below. Answer
what was asked — don't dump this entire document unprompted.

---

## The workflow in one picture

```
session start → /better-coding-start  (boot: load workflow + persist to constitution)
any task → /better-coding-orient  (route + steering)
              │
   ┌──────────┼──────────────┐
 trivial    small        medium/large
   │          │               │
   │     quick spec    /better-coding-plan
   │          │               │
   └────┬─────┴───────────────┘
        ▼
/better-coding-workflow   ←  + /better-coding-frontend (if UI)
        ▼
/better-coding-review     ←  + /better-coding-audit (if security-critical)
        ▼
/better-coding-verify  →  merge / PR / keep / discard

when something breaks at any point: /better-coding-debug
```

Each phase produces an artifact that constrains the next; rigor scales with
task size. That is what separates reliable agentic coding from vibe coding.

## The ten skills

| Skill | When | Modifies code? |
|-------|------|----------------|
| `/better-coding-start` | Once per session — boots the workflow, installs the durable constitution block | No (writes constitution after confirm) |
| `/better-coding-orient` | Every new, not-yet-scoped task — routes to the right process level, bootstraps project rules | No |
| `/better-coding-plan` | Non-trivial features/refactors — spec, design, task breakdown | No (docs only) |
| `/better-coding-workflow` | All implementation — the engineering discipline while writing code | Yes |
| `/better-coding-frontend` | Any UI work — a11y, UX states, i18n, privacy quality floor | Build mode only |
| `/better-coding-review` | Before merge — read-only two-stage review of a diff/PR | No |
| `/better-coding-audit` | Periodic/pre-release — whole-repo health check, 12 dimensions | No |
| `/better-coding-debug` | Anything broken — root-cause-first, no guess-and-check | Yes (the fix) |
| `/better-coding-verify` | Before claiming done — evidence gate + branch finishing | No |
| `/better-coding-help` | Questions about this skill set (this skill) | No |

## Common chains

- **Feature:** orient → plan → (approve) → workflow [+frontend] → review → verify
- **Bug:** debug → workflow (fix) → review → verify
- **Health check:** audit → plan (roadmap items) → workflow → review → verify
- **Trivial fix:** orient states "skipping planning — trivial change" → workflow

In plans, every work package starts with step 0: invoke
`/better-coding-workflow` (+ `/better-coding-frontend` for UI packages) so the
discipline is fresh in context before each package.

## Invoking the skills

- **Automatic** — each skill triggers on matching tasks via its description;
  you rarely need to name them.
- **Claude Code (manual)** — type the slash command, e.g. `/better-coding-plan`.
- **OpenAI Codex (manual)** — type `$` and pick the skill (e.g.
  `$better-coding-plan`), or run `/skills` to list what is installed.
- **Priority** — your `CLAUDE.md`/`AGENTS.md` instructions override skills;
  project-level skills override global ones of the same name.
- **When unsure** — if there is even a small chance a skill applies, load it;
  under-triggering costs more than a wasted load.

## Install / update / uninstall

The skills are plain folders with a `SKILL.md`. Claude Code scans
`.claude/skills/` (project) and `~/.claude/skills/` (global); Codex scans
`.agents/skills/` and `~/.agents/skills/`. Install by copying the folders in,
update by replacing them, uninstall by deleting them. The source repo's
`README.md` and `INSTALL.md` cover other agents and details.

## Two usage tips

- **Compact context around 50%.** Long sessions degrade output quality;
  compact or start fresh rather than pushing through.
- **CI owns the hard gates.** Skills make checks consistent, not unskippable —
  keep lint, type-check, security scan, and tests in CI.

## Hand-off

After answering, if the user has actual coding work, invoke
`/better-coding-orient` (or the directly matching skill) — this skill only
explains.