Bootstrap a repo

Bootstrap is how you turn a new repository into a first-class ClosedLoop.ai project. The bootstrap plugin reads your code and docs, identifies domains and languages, and generates agent prompts that know your code.

Prerequisites

• the six ClosedLoop plugins installed
• a project with at least a README.md; CLAUDE.md and ARCHITECTURE.md help a lot
• a sandbox that includes this repo

Run it

claude /bootstrap:start

Or:

claude /bootstrap:agent-bootstrap --depth medium --strategy backup

Useful flags

• --depth quick|medium|deep — how deeply to explore.
• --focus <area> — focus on one domain.
• --dry-run — plan without writing.
• --update — regenerate only agents whose project-context hash has changed.
• --strategy backup|skip|overwrite|interactive — conflict strategy (default backup).
• --add-domain <domain> — add a specific domain.

What it does

Nine phases:

1. Ingest CLAUDE.md, README, ARCHITECTURE.md, and manifests → discovery/project-context.md.
2. Detect languages by file count → discovery/languages.json (parallel with 3).
3. Identify domains → discovery/domains.json (parallel with 2).
4. Map domains and languages to candidate agents (always adds test-strategist and security-privacy) → synthesis/expert-agents.json.
5. Decompose complex agents into specialists; emit critic-gates.json → synthesis/decomposed-agents.json.
6. Pre-generation validation → synthesis/generation-validation.json.
7. Generate agents (fan-out, max 5 concurrent) → .claude/agents/<name>.md and .closedloop-ai/bootstrap-metadata.json.
8. Per-file validation → synthesis/agent-validation.json.
9. Final validation → validation-report.json and bootstrap-report.md.

What you get

• A set of .claude/agents/*.md tailored to your codebase, each with domain-appropriate colors, descriptions, and skill wiring.
• .closedloop-ai/settings/critic-gates.json that the code plugin uses to pick critics per task type.
• A persistent .closedloop-ai/bootstrap-metadata.json with SHA-256 hashes so --update can detect changes.

When to re-run

• When the project's architecture changes materially.
• When you add a new domain (--add-domain).
• When you want to refresh agents against new conventions (--update).
• On demand with --dry-run to compare what the generator would produce to the current agents.

Output location

.claude/agents/<agent-name>.md
.closedloop-ai/bootstrap-metadata.json
.closedloop-ai/settings/critic-gates.json
.closedloop-ai/bootstrap/<timestamp>/discovery/
.closedloop-ai/bootstrap/<timestamp>/synthesis/
.closedloop-ai/bootstrap/<timestamp>/validation-report.json
.closedloop-ai/bootstrap/<timestamp>/bootstrap-report.md

After bootstrap, run your first loop with /code:code to validate the agent suite against a real task.