create

name	create
description	Guided Claude Code setup — creates CLAUDE.md, settings.json, rules, and optional hooks/agents/skills for any project. Use when the user asks to initialize or set up Claude Code for a project.

Claude Code Project Setup

You are a Claude Code configuration expert. Set up optimal Claude Code configuration through interactive conversation.

Follow these phases in order.

---

Phase 0: Load Context & Learn

Read ../../references/learning-system.md and follow the Common Phase 0 steps (including Step 0.5 Migration & Stale Check) with these create-specific overrides:

• Step 2 override: When filtering local/recommendations.json, focus on issued_by in ["secure", "optimize"] (to avoid overwriting changes those skills applied). This is declared in learning-system.md as a create-specific cross-skill override.

After completing Common Phase 0:

• From /create-issued recommendations in recommendations.json: note previously DECLINED features. Do NOT re-suggest declined features in Phase 2A Question 6 unless the user asks.

Phase 0.5: Determine Path

If profile.json was loaded in Phase 0: The project type is already known. Use the profile's project_structure field to determine if this is an existing project. If the profile's claude_code_configuration_state indicates claude_md and settings_json are present, route directly to the Incremental path (Phase 2A-Incremental) without asking. Otherwise, skip to the "existing vs new" question below.

First, silently check if Claude Code configuration already exists:

1. Check for CLAUDE.md (root or .claude/CLAUDE.md)
2. Check for .claude/settings.json
3. Check for .claude/rules/ directory

If both CLAUDE.md and settings.json exist → ask:

"I see you already have Claude Code configuration set up. What would you like to do?"

(a) Add missing features — keep existing config, add what's missing (rules, hooks, agents, skills, security) (b) Start fresh — regenerate configuration from scratch (existing files will be merged, not overwritten)

If the user chooses (a), read references/best-practices.md then templates/advanced.md and follow the Incremental path (Phase 2A-Incremental).

If the user chooses (b), continue to the question below.

If no existing config (or user chose "start fresh") → ask:

"Is this an existing project with code, or a new/empty project you're starting from scratch?"

(a) Existing project — I already have code, dependencies, and/or a framework set up (b) New/empty project — I'm starting from scratch or have only basic scaffolding

---

Route to Path

Based on the user's choice, read the generation rules and the appropriate path file (Phases 1–3 are defined in each path file):

If the user chooses (b) New/empty project:

1. Read references/best-practices.md — common generation rules
2. Read templates/starter.md — follow the Starter path instructions

If the user chooses (a) Existing project:

1. Read references/best-practices.md — common generation rules
2. Read templates/advanced.md — follow the Advanced path instructions (references ../../references/security-patterns.md for security configuration)

After completing all generation steps from the path file, return here for Phase 4.

---

Contract Framework

The generated configuration should incorporate four agent-contract patterns. Each is documented in a guide that the path files (templates/starter.md, templates/advanced.md) embed when generating content:

• Identity-DNA (CLAUDE.md role declaration, mental models for user and agent, deliverable invariant) — see docs/guides/claude-md-guide.md § Identity-DNA
• Memory patterns (four memory types, frontmatter schema, MEMORY.md index, what-NOT-to-save, verify-before-recommending) — see docs/guides/memory-patterns-guide.md
• Three permission tiers (settings.json allow / ask / deny mapped to Regular / Explicit-permission / Prohibited) — see docs/guides/settings-guide.md § The three permission tiers
• Verification discipline (read-back-after-edit, tool-success ≠ task-correct, scope-checked reporting) — see plugin/references/verification-discipline.md

When Phase 3.5 Quick Verify runs, the four patterns above should be discoverable in the generated CLAUDE.md and settings.json (or explicitly skipped per the user's Phase 2A answers). Path files embed contract-aware content directly; this overview lets future skill maintenance verify that generated output still includes the four patterns.

---

Phase 3.5: Quick Verify

After generating files, run a quick sanity check before wrapping up:

1. CLAUDE.md — confirm it has at least: Project Overview, Build & Run, Testing sections
2. settings.json — confirm it has valid permissions with both allow and deny arrays
3. Rule files (if created) — confirm each has YAML frontmatter with at least a # heading
4. Agent files (if created) — confirm each has model: in frontmatter and a Scope section
5. Cross-references — confirm any directories mentioned in CLAUDE.md actually exist in the project
6. Command dependencies — confirm every suggested command only references packages listed in the project's dependency manifest (e.g., if pytest --cov is suggested, verify pytest-cov is in requirements.txt; if eslint is suggested, verify it's in package.json)

If any check fails, fix it immediately before proceeding. Do not ask the user — just correct the issue and note it in the Phase 4 summary.

---

Phase 4: Wrap Up

After generating all files:

1. Print a summary table listing every created/modified file and what it contains

2. If you merged into existing files, explain what was added

3. Tell the user: "Run /memory to verify all configuration files are loaded"

4. Suggest trying a simple task to test the configuration works

5. Suggest: "Run /guardians-of-the-claude:audit to evaluate your new configuration."

If the user followed the Starter path, also add:

You're using a starter configuration. As your project grows and you want rule files, hooks, agents, or skills, run /guardians-of-the-claude:create again and choose "Existing project" to upgrade to the full configuration.

Phase 4.5: Persist Results & Learn

Read ../../references/learning-system.md and follow the Common Final Phase steps with these create-specific overrides:

• Step 1 override (Skill-specific data in changelog entry): The config-changelog.md entry for this skill must include:

  • Detected: — include the path taken (starter, advanced, or incremental) so future runs can reason about what scaffold the project started from.
  • Applied: — list of files created or modified this run (CLAUDE.md, settings.json, rules/, agents/, hooks, skills, MCP config).
  • Recommendations: — features the user explicitly skipped in Phase 2A Question 6 marked as DECLINED by user. This enables Learning Rule 2 (Preference Respect) in future runs.

  Each DECLINED entry written this run must increment decline_count per plugin/references/lib/merge_rules.md §recommendations.json merge rules: if the rec is a fresh DECLINE (no current entry, or current entry was PENDING), set decline_count = 1; if re-recording a DECLINE on an already-DECLINED rec, increment decline_count++ (monotonic — never decremented even when a previously-DECLINED rec re-transitions to PENDING). Writes always emit schema 1.1.0; reading a 1.0.0 file performs lazy migration in-memory (inflate missing decline_count to 0) before the merge.

  Profile merge (lock-free in Final Phase Step B, against the Step A snapshot): /create owns the project-structure sections (runtime_and_language, framework_and_libraries, package_management, testing, build_and_dev, project_structure) and claude_code_configuration_state.claude_md. It must update claude_code_configuration_state.{rules_count, agents_count, hooks_count, mcp_servers_count} for entity classes it added (see plugin/references/lib/merge_rules.md §profile.json merge rules). Sections owned exclusively by /secure (settings_json) must be preserved from snap_profile (the profile.json captured in the Step A snapshot).

  Sections owned exclusively by /audit (monorepo_detection) must be preserved. Per plugin/references/lib/merge_rules.md §project_structure / monorepo_detection consistency precondition:

  current_profile.monorepo_detection.detected	project_structure.type rule for /create
  true	preserve "monorepo"
  false	preserve existing value if present; when writing project_structure.type, write "single_project" only (must NOT write "monorepo" or null)
  null (or monorepo_detection == null)	may write "single_project" or null (must NOT write "monorepo")

  Never write monorepo_detection itself — /audit is exclusive owner.

  A1 merge rule amendments (applied summary; mechanism in plugin/references/lib/merge_rules.md):

  • Row 1 — claude_code_configuration_state.model: any-skill writer; last-write-wins; written at Step 0.5 and Final Phase. Stateless mode: no-op.
  • Row 3 — config-changelog.md entry - Model: bullet: /create delta-emits per the shared hybrid writer policy. See plugin/references/learning-system.md § Model Bullet Emission for full mechanics; this skill emits only when the resolved model differs from the immediately previous entry's bullet value.

After completing Common Final Phase, run Critical Thinking & Insight Delivery from the learning system reference.