Coding setup

Coding Setup

This skill sets up Claude Code for professional software development using a proven iterative workflow.

Handoff context: Read skills/_shared/consume-handoff.md and run it with the handoff block (if any). The helper guarantees the following locals: detected_language, existing_claude_md, inferred_use_case, repo_signals, graphify_candidate. Use detected_language for all user-facing prose; generated file content stays in English.

Existing CLAUDE.md: If existing_claude_md: true, DO NOT overwrite it. Append a new delimited section at the end of the file:

<!-- onboarding-agent:start setup=coding skill=coding-setup section=claude-md -->
## Claude Onboarding Agent — Coding Setup
...generated content...
<!-- onboarding-agent:end -->

If the delimited block already exists from a previous run, replace only the content between the markers; leave the rest of the file untouched. Use the same marker pattern (# onboarding-agent: coding — start / — end) around generated .gitignore blocks so /upgrade-setup can refresh them non-destructively.

Supporting Files

Read these on-demand at the step that invokes them. Do not read eagerly.

• skills/_shared/consume-handoff.md — orchestrator handoff parse + inline fallback (preamble, before Step 1)
• skills/_shared/offer-superpowers.md — canonical Superpowers opt-in and install (Step 1)
• skills/_shared/offer-github-mcp.md — canonical GitHub MCP offer (Step 4)
• skills/_shared/offer-graphify.md — canonical Graphify opt-in (Step 7)
• skills/_shared/gitignore-python.md, skills/_shared/gitignore-node.md, skills/_shared/gitignore-common.md — canonical .gitignore blocks (Step 5)

Step 1: Install Dependencies

Read skills/_shared/offer-superpowers.md and run it with skill_slug: coding-setup, mandatory: true. The helper delegates to skills/_shared/installation-protocol.md and sets superpowers_installed, superpowers_scope, superpowers_method.

Step 2: Context Questions

Ask these 2 questions one at a time. Wait for each answer before asking the next:

1. "What is your primary programming language or tech stack? (e.g., Python, TypeScript/React, Go, Rust)"
2. "Which IDE or editor do you use? (VS Code / JetBrains / Neovim / other)"

Step 3: Offer Project-Local Subagent

Read skills/_shared/emit-subagent.md and follow it with these inputs:

• slug: code-reviewer

• purpose_blurb: "Review a PR-sized diff (uncommitted, staged, or a named commit range) against the project's code standards."

• frontmatter_description: "Use to review uncommitted diffs, staged changes, or a named commit range against the project's code standards and conventions. Dispatch when the user asks for a code review, wants feedback on a change, or says 'review this' / 'check my changes' / 'review my diff'."

• tools_list: Bash, Read, Grep, Glob

• rules_files: (none — this subagent relies on CLAUDE.md's code standards)

• body_markdown:

  You are the Code Reviewer. You review a bounded diff against the project's code standards and conventions documented in CLAUDE.md.
  
  ## Procedure
  1. Determine the diff scope from the caller's request (uncommitted changes, staged only, a named commit range, or a specific file list). If ambiguous, ask once; otherwise proceed with `git diff` on uncommitted + staged.
  2. Read CLAUDE.md for project context and standards.
  3. Review the diff. Flag: YAGNI violations, speculative abstractions, missing validation at boundaries, error-handling for impossible scenarios, test coverage gaps, convention drift.
  4. Return a structured verdict: one-line summary, bullet list of findings (severity: blocker / suggestion / nit), and a final recommendation (ship / revise / block).
  
  ## Rules
  - Do not write code. Your output is feedback, not a fix.
  - Do not run tests or linters on the caller's behalf unless explicitly asked.
  - Cite file:line for every finding.
  

Record the emit outcome (emit_subagent, subagent_skipped_existing, subagent_deferred) for use in the completion summary (Step 10). If emit_subagent: true, add "code-reviewer" to the list passed to skills/_shared/write-meta.md in Step 8 as subagents_installed.

Step 4: Offer GitHub MCP (conditional)

Read skills/_shared/offer-github-mcp.md and run it with skill_slug: coding-setup. The helper carries the canonical trigger condition, install command, auth details, and pointer link, and sets github_installed (and github_deferred on later) for the CLAUDE.md generator and completion summary.

Step 5: Generate Artifacts

Generate the following files automatically:

CLAUDE.md

# Claude Instructions

## Project Context
[Stack from Q1] project. IDE: [Q2 answer].[If `git remote -v` reveals a recognizable host (github.com / gitlab.com / bitbucket.org), append ` Git host: <host>.`; otherwise omit the sentence.]

## Development Workflow
[Include this entire section ONLY if superpowers_installed is true]

At the start of every new conversation, invoke the `using-superpowers` skill.

Follow this iterative workflow for all features and bug fixes:
1. **Brainstorm** — use `superpowers:brainstorming` to explore requirements before writing any code
2. **Plan** — use `superpowers:writing-plans` to create a step-by-step implementation plan
3. **Implement** — use `superpowers:subagent-driven-development` to dispatch subagents per task
4. **Review** — use `superpowers:requesting-code-review` before merging
5. **Commit** — atomic commits with descriptive messages after each verified task

## Code Standards
- Write minimal code. YAGNI: do not add features not explicitly requested.
- No speculative abstractions. Three similar lines beat a premature helper.
- Validate only at system boundaries (user input, external APIs).
- No error handling for scenarios that cannot happen.

## Testing
- Write tests before implementation (TDD).
- Tests must run against real dependencies — avoid mocks unless unavoidable.

[Include ONLY if github_installed is true OR github_deferred is true — emitted per skills/_shared/offer-mcp.md Step 5]
## Configured MCP servers
- github: [see _shared/offer-mcp.md Step 5 for the exact per-state line format]

Adapt the "Project Context" section based on the detected stack:

• Python → mention uv as package manager if applicable
• Node/TypeScript → mention npm/yarn/pnpm as appropriate
• Go/Rust → note the standard build tools

AGENTS.md

# Agent Roles

## coder
Implements features and bug fixes. Works from a written plan. Makes atomic commits after each verified task. Does not refactor code unrelated to the current task.

## reviewer
Reviews code changes against the implementation plan and coding standards. Checks for correctness, security (OWASP Top 10), and spec alignment. Returns a list of issues with severity (critical/minor).

## git-agent
Handles all git operations: staging, committing, pushing. Writes descriptive commit messages. Never force-pushes to main. Confirms before any destructive git operation.

.claude/settings.json

Create .claude/settings.json with stack-appropriate tool permissions:

{
  "permissions": {
    "allow": [
      "Bash(git *)"
    ]
  }
}

Extend the allow list based on detected stack:

• Python → add "Bash(uv *)", "Bash(python *)", "Bash(pytest *)"
• Node/TypeScript → add "Bash(npm *)", "Bash(node *)", "Bash(npx *)"
• Go → add "Bash(go *)"
• Rust → add "Bash(cargo *)"

.gitignore

Generate a .gitignore appropriate for the detected stack. Always include the common block from skills/_shared/gitignore-common.md, then append the stack-specific block:

• Python → read skills/_shared/gitignore-python.md and append its block
• Node / TypeScript → read skills/_shared/gitignore-node.md and append its block
• Go → append *.exe, *.test, vendor/
• Rust → append target/
• Generic fallback → nothing beyond the common block

Wrap the generated block in # onboarding-agent: coding — start / — end markers so /upgrade-setup can refresh it non-destructively.

Step 6: Optional Community Skills

"Would you like to install additional community skills?

A) frontend-design (official Anthropic) — avoids AI-generic UI, bold design decisions (popular official plugin) B) mcp-builder — create MCP servers for external API integrations C) webapp-testing — Playwright-based UI testing D) security-suite — Trail of Bits CodeQL/Semgrep analysis E) All of the above F) None

(Multiple selections via comma, e.g. 'A, C')"

For each selected skill, run: /plugin install <skill>@claude-plugins-official

On failure for any skill: warn clearly ("⚠ Could not install [skill] — skipping. Install manually later.") and continue. Never block the setup.

Add the list of successfully installed optional skills to the Completion Summary under a new line: Optional community skills: [list or "none selected"]

Step 7: Optional Graphify Integration

Read skills/_shared/offer-graphify.md and run it with:

• host_setup_slug: "coding"
• host_skill_slug: "coding-setup"
• run_initial_build: true
• install_git_hook: true
• corpus_blurb: "your project (code via tree-sitter for 25 languages, plus Markdown, PDFs, diagrams, images, audio/video)"

The helper owns the opt-in prompt and the three-way branch (yes / no / later), delegating to skills/_shared/graphify-install.md. Record the graphify_* variables it produces for use in Step 10.

Step 8: Write Upgrade Metadata

Set setup_slug: coding, skill_slug: coding-setup. Resolve plugin_version from the plugin's own plugin.json. If Step 3 emitted the code-reviewer subagent, set subagents_installed: ["code-reviewer"]; otherwise leave it unset (the helper treats it as an empty list). Then follow skills/_shared/write-meta.md to create or merge ./.claude/onboarding-meta.json. If Step 7 installed Graphify, skills_used will automatically pick up graphify-setup via the shared protocol's own write-meta call — this step records coding-setup alongside it.

Step 9: Render Anchor Sections

Read skills/_shared/anchor-mapping.md. Locate the row for setup_type: coding. For each anchor slug in that row:

1. Call skills/_shared/render-anchor-section.md with:
   • setup_type: coding
   • skill_slug: coding-setup
   • anchor_slug: <slug>
   • target_file: ./CLAUDE.md
   • fallback_content: <embedded fallback from skills/anchors/SKILL.md for that slug>
2. If a ./AGENTS.md file was generated earlier in this skill, repeat the call with target_file: ./AGENTS.md.

Do not fail if any single render-anchor-section.md call returns placeholder — that is the designed offline path. Collect the list of rendered / placeholder slugs to mention in the completion summary.

For every call, also capture render_freshness. When it is anything other than network or cache (i.e. fallback or embedded), record the (anchor_slug, render_freshness) pair in anchor_freshness_notes. The completion summary's Anchor freshness line consumes this list.

Step 10: Completion Summary

✓ Coding setup complete! Here's what was configured:

Files created:
  CLAUDE.md                     — project context + [workflow instructions / ⚠ omitted — Superpowers not installed]
  AGENTS.md                     — coder, reviewer, and git-agent role definitions
  .claude/settings.json         — tool permissions for [stack]
  .gitignore                    — [stack]-appropriate ignore rules
  .claude/agents/code-reviewer.md — project-local subagent (auto-invoked) [only on yes path; if skipped existing: .claude/agents/code-reviewer.md (already existed — skipped; re-run /checkup --rebuild to regenerate); if no/later: Subagent code-reviewer not installed — re-run /coding-setup to add it later.]
  .claude/onboarding-meta.json  — setup marker for /upgrade-setup

External skills:
  [✓ Superpowers installed via superpowers_method (superpowers_scope)]
  [⚠ Superpowers installation failed — install manually: https://github.com/obra/superpowers]

Optional community skills: [list of installed skills, or "none selected"]

Graphify (knowledge graph):
  [✓ installed via <installer>, /graphify + PreToolUse hook registered | ⚠ installed but hook not verified — run /graphify in a new session | — skipped: <reason> | — deferred: run /graphify-setup when ready | — not offered]

MCP servers:
  [one line per MCP considered, formatted per skills/_shared/offer-mcp.md Step 6 — omit if github trigger condition was false]

Anchor freshness:
  [omit the whole block if anchor_freshness_notes is empty; otherwise one line per entry:
   Anchor <anchor_slug> served from <render_freshness> — consider running /anchors to refresh.]

Next steps:
  Start a new Claude session and run: /superpowers:using-superpowers
  Then try: /superpowers:brainstorming "describe your first feature"
  [If Graphify installed] Try: /graphify query "where does auth happen in this repo?"
  - Run `/anchors` any time to refresh the anchor-derived sections. If any section was rendered as a placeholder due to offline mode, re-run `/anchors` once you are back online.