Skip to content

INSTALL-WITH-CLAUDE.md

Latest commit

 

History

History
139 lines (104 loc) · 6.29 KB

INSTALL-WITH-CLAUDE.md

File metadata and controls

139 lines (104 loc) · 6.29 KB

Guided Installation via Claude Code

Run this prompt in a Claude Code session to have Claude guide you through the full installation and configuration process interactively.

Usage

cd /path/to/hivedriver
claude -p < INSTALL-WITH-CLAUDE.md

Or paste the prompt below into an interactive Claude Code session.

Prompt

You are helping the user install and configure the DCS Hive-Mind system for Claude Code. This is an interactive setup — ask questions, make suggestions, and populate configuration files based on the user's answers. Be conversational and helpful.

Phase 1: Run the Installer

  1. Read the README.md in this directory to understand what is being installed.
  2. Run ./install.sh and guide the user through the prompts:
    • If multiple Claude Code installations are found, explain the difference and recommend the global one.
    • Explain the two install modes:
      • Replace is best for fresh setups or if they want DCS as their primary workflow.
      • Additive is best if they already have custom agents they want to keep.
    • If conflicts are detected in additive mode, reassure the user that their existing agents are preserved and explain the hive- prefix.
  3. Confirm the installer completed successfully by checking that ~/.claude/skills/hive-mind/SKILL.md exists.

Phase 2: Configure Identity and Environment

Read the installed ~/.claude/hive-mind.env file. For each variable, ask the user and update the file:

  1. USER_NAME: "What name should I use when addressing you in design documents?"

    • This appears in agent prompts like "Return results to {USER_NAME}".

  2. NOTIFY_EMAIL: "Would you like to receive an email when a hive-mind design completes? If so, what email address?"

    • If yes, also ask about their notification method:
      • Do they have a mail script/CLI tool? (e.g., msmtp, sendmail, a custom script)
      • A webhook? (Slack, Discord, etc.)
      • Set HIVE_MIND_NOTIFY_CMD accordingly.
    • If no, leave both blank and explain the design will just be available in OUTPUT_DIR.

  3. Deployment targets: "Do you have dev/staging/prod environments where you deploy applications?"

    • For each environment they have, ask for user@host:port format.
    • Explain this is used by the test-lead agent for deployment verification health checks.
    • It's fine to leave these blank — they can add them later.

  4. Project classification: "Do you have any high-security projects that should always get a deep security review?"

    • Explain: HIGH_SECURITY projects get mandatory security-analyst review with Gemini cross-validation.
    • Ask for comma-separated project names.
    • Also ask about compliance frameworks if they handle PII, health data, financial data, etc.
    • Suggest appropriate frameworks: privacy-act-au, gdpr, hipaa, pci-dss, sox.

  5. Model preferences:

    • "Are you using Claude Max subscription or API billing?"
      • If subscription: leave budgets as-is (they're caps, not costs).
      • If API: ask if they want to adjust ARCHITECT_BUDGET and CHALLENGER_BUDGET.
    • "Do you have Gemini CLI installed and authenticated?"
      • Test with: gemini -p "Say OK" -o text 2>/dev/null
      • If it works: leave GEMINI_ENABLED=true.
      • If not: set GEMINI_ENABLED=false and explain they can enable it later.

After each answer, update the ~/.claude/hive-mind.env file using the Edit tool.

Phase 3: Configure User Policies

Read ~/.claude/skills/hive-mind/user-config.md. Explain what it's for, then ask the user about their standing preferences:

  1. Technology constraints: "Are there any technologies or frameworks you want to avoid or prefer for new projects?"

    • Example: "I'm evaluating Svelte over React for the next 3 months"
    • Example: "Always use Python 3.12+ for new backend services"

  2. Design philosophy: "Any strong opinions on design patterns or architecture?"

    • Example: "Prefer composition over inheritance"
    • Example: "Microservices for new projects, but no Kubernetes"

  3. Code quality standards: "Any specific testing or quality bars?"

    • Example: "Minimum 80% test coverage"
    • Example: "All public APIs must have OpenAPI specs"

  4. Deployment rules: "Any restrictions on when or how things get deployed?"

    • Example: "No production deploys on Fridays"
    • Example: "All deploys need a rollback plan documented"

  5. Other constraints: "Anything else the design and implementation agents should know about?"

    • Accessibility requirements, performance budgets, licensing restrictions, etc.

For each policy the user provides, write it to ~/.claude/skills/hive-mind/user-config.md in the structured format:

## [Policy Name]
**Effective:** [date range or "permanent"]
**Rule:** [the constraint]
**Rationale:** [why this matters]
**Escape hatch:** [when it's OK to deviate, if ever]

If the user doesn't have policies for a category, skip it — an empty config is fine.

Phase 4: Verification

  1. Confirm all files are in place:

    ls -la ~/.claude/hive-mind.env
ls -la ~/.claude/skills/hive-mind/SKILL.md
ls -la ~/.claude/skills/hive-mind/pipeline.md
ls -la ~/.claude/skills/hive-mind/user-config.md
ls -la ~/.claude/agents/architect.md  # or hive-architect.md in additive mode

  2. Verify settings.json has the agent teams env var:

    python3 -c "import json; d=json.load(open('$HOME/.claude/settings.json')); print(d.get('env',{}).get('CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS','NOT SET'))"

  3. Show the user a summary of what was configured:

    • Their name and notification setup
    • Number of agents installed (and whether any were namespaced)
    • Deployment targets configured
    • Security classifications
    • User policies defined

  4. Explain how to use the system:

    • /hive-mind [n] [prompt] for multi-perspective design
    • After design: reference ~/.claude/skills/hive-mind/pipeline.md for implementation
    • Edit configs anytime — they're re-read on each run

  5. Suggest a test run: "Want to try a quick test? Run: /hive-mind Design a simple health check endpoint"

Tone

Be conversational and explain things clearly. Don't overwhelm with options — if the user seems unsure, suggest sensible defaults. It's always fine to skip optional configuration and come back later. The goal is a working setup, not a perfect one.