Bootstrap agent

Bootstrap Agent

One-time interactive wizard that configures this template repo into a working AI agent. After running, the repo will have a complete CLAUDE.md, GitHub repo, CI/CD, and skill-manager configured.

---

Workflow

Before Starting

Confirm with the user: "This will configure the template into a new AI agent repo. It's a one-time process — ready to go?"

Step 1: Identity

Infer the agent name from the current repository directory name. If the directory ends with -ai, strip the suffix (e.g. mybot-ai → mybot). Present the inferred name and ask the user to confirm or provide a different name.

If the name cannot be inferred (e.g. the directory is still called ai-agent-template), ask the user to provide one.

Validation: The name must be a single word, lowercase, alphanumeric + hyphens only (^[a-z][a-z0-9-]*$).

Then ask these questions one at a time:

1. Purpose — One sentence describing what this agent does (e.g. "An AI assistant that manages daily task reviews and email triage").
2. Persona — How should the agent communicate? What's the tone? (e.g. "Professional but approachable. Uses clear, direct language.")

Step 2: MCP Dependencies

Present common MCP server categories and ask which ones this agent needs:

Category	Example Use Cases
Calendar	Event management, scheduling, availability checks
Task Management	Todo lists, project tracking, inbox processing
Browser Automation	Web interaction, form filling, page scraping
Database / Storage	Persistent data, logging, document management
Communication	Email, messaging, notifications

Ask:

• "Which of these categories does your agent need?"
• "Do you have specific MCP servers in mind, or should we leave placeholders?"

Build the MCP table for CLAUDE.md from the answers.

Step 3: Data Storage

Ask the user:

• "Does this agent need to store data files?" (e.g. JSON records, markdown logs, session data)
• If yes: "What format? (JSON, markdown, other) What directory should they live in?"
• "Does this agent need an inputs directory?" (for files the user drops in for processing)

Create any needed directories and update .gitignore accordingly.

Step 4: Scheduled Tasks

Ask the user:

• "Will this agent have any scheduled tasks?"
• If yes, for each task gather:
  • Name (kebab-case, e.g. daily-morning-scan)
  • Cron expression (e.g. 0 10 * * 1-5 for weekdays at 10:00)
  • Model (default: Claude Opus 4.6)
  • Brief description of what it does

Create placeholder files in scheduled-tasks/ with metadata tables.

Step 5: Commit Scopes

Present the default scopes: skills, config, ci, docs, evals, scripts

Ask: "Does this agent need any additional commit scopes?" (e.g. data, profiles, workflows)

Step 6: Agent Configuration

Ask the user:

• "What business rules, categories, or domain-specific configuration should go in the CLAUDE.md?"
• "This is the heart of the agent — it's where all the rules and logic live. We can flesh this out now or leave placeholders and iterate later."

If the user wants to do it now, help build out the configuration section. If not, leave clear <!-- TODO: --> markers.

Step 7: Generate Files

Replace all {{placeholders}} across the repo:

Files to update:

• .claude/CLAUDE.md — Replace {{AGENT_NAME}}, {{AGENT_PURPOSE}}, {{AGENT_PERSONA}}, {{MCP_TABLE}}, {{SCOPES_LIST}}, {{AGENT_SPECIFIC_CONFIG}}
• .commitlintrc.json — Add any extra scopes from Step 5
• README.md — Replace {{AGENT_NAME}}, {{AGENT_PURPOSE}}, {{SKILL_ROWS}}, {{TASK_ROWS}}, {{MCP_ROWS}}
• CONTRIBUTING.md — Replace {{AGENT_NAME}}, {{SCOPES_LIST}}
• skills-manifest.json — Regenerate with scripts/generate-manifest.sh

After replacing placeholders, delete the .claude/skills/bootstrap-agent/ directory — this is a one-time skill and is not needed after bootstrap completes.

Regenerate the manifest: scripts/generate-manifest.sh

Step 8: Git — Safety Check, Commit & Push

First, check the git remote:

git remote get-url origin

If the remote URL matches the template repository (e.g. ends with /ai-agent-template or /ai-agent-template.git):

STOP. The user has cloned the template repository directly. Pushing would target the upstream template. Warn the user and suggest:

1. Fork the repo on GitHub — creates a personal copy with upstream connection for future template updates
2. Use the template — run gh repo create <name> --template <template-url> to create an independent copy
3. Remove the remote manually — git remote remove origin, then set up a new remote

Do NOT proceed with any push until this is resolved.

If the remote points to the user's own repo:

Safe to proceed:

lefthook install
git add .
git commit -m "feat: bootstrap {name}-ai agent"
git push

Where {name} is the agent name from Step 1 (e.g. feat: bootstrap mybot-ai agent).

If there is no remote configured:

Ask the user if they'd like to create a GitHub repository:

gh repo create <owner>/<name> --private --source=. --push

Confirm the owner/org and visibility before executing.

Completion

Summarise what was created:

• Agent name and purpose
• Skills installed (any created during setup)
• Scheduled tasks configured
• MCP dependencies
• Repository URL (if pushed)
• Next steps: "Add your first skill, flesh out the CLAUDE.md config, or set up scheduled tasks"

---

Important Notes

• This skill runs once. After bootstrap, the bootstrap-agent/ directory is deleted.
• Always confirm before running destructive or external commands (git push, gh repo create).
• If gh CLI is not authenticated, tell the user to run gh auth login first.
• All generated files should pass lefthook run pre-commit before the initial commit.