Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root. If shared/ is missing, fetch files via WebFetch from https://raw.githubusercontent.com/levnikolaevich/claude-code-skills/master/{path}.

Dev Environment Setup

Type: L2 Domain Coordinator Category: 0XX Shared

Single-pass coordinator that scans, installs, and configures the full agent development environment: CLI agents, MCP servers, config sync across agents, hooks, and best practices audit. Delegates to 4 specialized workers.

When to Use This Skill

• First-time project setup (no docs/environment_state.json)
• After installing/removing CLI agents (Codex, Gemini)
• After adding/removing MCP servers in Claude Code
• When hooks need reconfiguration or verification
• Periodic alignment check across all agents

Input Parameters

Workflow

OS Detect → Full Scan → Delegate Workers → Best Practices Audit → Write State + Report

Phase 0: OS & Environment Detection

Config paths by OS:

Load disabled flags: Read docs/environment_state.json if exists. Extract agents.{name}.disabled for each agent.

Phase 1: Full Scan

Run all sub-checks before any modifications. Display scan table after.

1a: Probe CLI Agents

• node shared/agents/agent_runner.mjs --health-check — probes all registered agents
• Parse output: per-agent available, version
• If agent_runner.mjs not found: set both available: false

1b: Discover MCP Servers

• Run claude mcp list — canonical source of truth for configured servers
• Parse output: server name, transport type (stdio/HTTP), connection status
• Fallback if claude CLI unavailable: read ~/.claude.json + ~/.claude/settings.json, merge by server name

1c: MCP Token Budget

• Formula: server_count * 5000 tokens
• Context percentage: estimated_tokens / 200000 * 100
• Threshold: <=5 servers OK, 6-8 WARN, >8 WARN critical

1d: Hook Health

MANDATORY READ: Load shared/references/hook_health_check.md

• Validate hooks from both locations: hooks/hooks.json (plugin) and .claude/settings.local.json (project)
• Check JSON syntax, script existence, dependencies

1e: CLAUDE.md Health

• Exists in project root
• Line count (<100 recommended)
• No timestamps (breaks prompt cache)
• Has compact instructions section

1f: Target Agent Configs

• Per target (Gemini/Codex): config exists, MCP servers count, symlinks state

Display scan table:

Environment Scan:
| Area          | Status | Detail                        |
|---------------|--------|-------------------------------|
| Codex CLI     | ok     | 0.1.2503                      |
| Gemini CLI    | ok     | 2025.6.15                     |
| MCP servers   | ok     | 5 servers, 25K tokens (12.5%) |
| Hooks         | ok     | 3 events, 3/3 scripts found   |
| CLAUDE.md     | ok     | 85 lines, compact, no TS      |
| Gemini config | drift  | 3/5 servers synced            |
| Codex config  | ok     | 5/5 servers synced            |

Phase 2: Delegate to Workers (sequential, same context)

MANDATORY: All 4 workers MUST be invoked via Skill tool sequentially in this context. DO NOT skip any worker. DO NOT execute worker tasks inline — each worker MUST be invoked via Skill tool. Pass scan results as delegation context. If agent disabled: true, pass flag so workers skip operations for that agent.

Invocation (each step):

Skill(skill: "{worker}", args: "{args}")

Phase 3: Best Practices Audit

Aggregate checks from workers + coordinator's own checks into one table.

Phase 4: Write State + Report

1. Read existing docs/environment_state.json (preserve disabled flags)
2. Build new state validated against references/environment_state_schema.json:
   • scanned_at: ISO 8601 timestamp
   • agents: probe results merged with preserved disabled
   • claude_md: exists, line_count, has_timestamps, has_compact_instructions
   • config_sync: per-target synced status
   • best_practices: score and findings
   • NOT persisted (scan-time only): MCP servers, MCP budget, hooks — shown in scan table but not written to file (native CC settings are source of truth)
3. Ensure docs/ directory exists
4. Ensure gitignore: If inside a git repo (git rev-parse --git-dir), check .gitignore for docs/environment_state.json. If not covered → append:
   Copy

   # Machine-specific environment state (generated by ln-010)
   docs/environment_state.json
   

5. Write docs/environment_state.json (2-space indent)
6. Summary report:

Dev Environment Setup Complete:
| Area          | Result   | Actions Taken               |
|---------------|----------|-----------------------------|
| CLI agents    | ok       | Gemini updated to 2025.6.15 |
| MCP servers   | ok       | 5 configured, budget OK     |
| Config sync   | ok       | Gemini: 2 new, Codex: skip  |
| Hooks         | ok       | output-filter added         |
| Best practices| 9/10     | 1 WARN (see above)          |

State: docs/environment_state.json

Worker Invocation (MANDATORY)

All workers: Invoke via Skill tool sequentially — workers see coordinator context.

TodoWrite format (mandatory):

- OS & environment detection (pending)
- Full scan (pending)
- Invoke ln-011-agent-installer (pending)
- Invoke ln-012-mcp-configurator (pending)
- Invoke ln-013-config-syncer (pending)
- Invoke ln-014-agent-instructions-auditor (pending)
- Best practices audit (pending)
- Write state + report (pending)

Critical Rules

Anti-Patterns

Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

Skill type: execution-orchestrator. Analyze this session per protocol §7. Output per protocol format.

Definition of Done

• OS and environment detected (Phase 0)
• Full scan completed with table displayed (Phase 1)
• All 4 workers invoked with scan context (Phase 2)
• Best practices audit table shown with 9 checks (Phase 3)
• docs/environment_state.json written (Phase 4)
• docs/environment_state.json covered in .gitignore (or skipped if not a git repo)
• Existing disabled flags preserved across rescan
• Summary report displayed to user

Version: 1.0.0 Last Updated: 2026-03-20