kaddo bootstrap

kaddo bootstrap

kaddo bootstrap creates the structural knowledge baseline Kaddo expects. It is not “new-project bootstrap” — it is knowledge-baseline bootstrap and applies to every project type. What changes by project.state is the template content and orientation, not whether bootstrap applies:

State	Orientation
new	intent — product vision, planned capabilities, initial technical direction
pre-ai	discovery — current state, observed capabilities, assumptions, open questions
legacy	risk — constraints, criticality, dependencies, technical debt, modernization

bootstrap is deterministic: it never calls an LLM, never generates source code, never decides the architecture, never installs agents/skills, and never runs scan/context/git. It writes starter templates (with placeholders, assumptions and [open] questions) that you then refine with agents.

What it creates

The common baseline (files + directories), with state-aware content:

knowledge/business/business.md
knowledge/product/product.md
knowledge/product/capabilities.md
knowledge/tech/codebase.md
knowledge/tech/current-state.md
knowledge/tech/decisions/
knowledge/delivery/roadmap.md
knowledge/delivery/work-items/

Every generated file carries project_state: in its front matter. pre-ai and legacy templates add discovery/risk sections (e.g. Observed technical signals, Risks of interpretation, Critical dependencies, Modernization notes).

Existing capability discovery (pre-ai / legacy)

For existing projects, knowledge/product/capabilities.md is scaffolded as an evidence-backed capability inventory, not a wishlist. The capability-agent fills it in a state-aware way:

• new → Planned Capability Definition ([planned] capabilities).
• pre-ai → Existing Capability Discovery — a ## Capability Domains map where capabilities are grouped by functional domain (### Domain: <name> with a Purpose + Evidence summary), each #### Capability: carrying a Status (implemented/partial/inferred/risky/deprecated/unknown) with evidence (paths, routes, tables, functions), plus ## Capability Gaps and ## Roadmap Candidate Signals (each naming its Domain + Related capability).
• legacy → Legacy Capability Discovery — the same domain map plus Criticality, Change risk, Operational dependency and Modernization notes per domain/capability.

Domains are grouped by functional responsibility (Loyalty, Billing & Subscriptions, …) — never by technical folder (src/components, src/app/api).

The roadmap-agent then treats capabilities.md as its primary source for roadmap candidates (partial capabilities, gaps, candidate signals, risky capabilities) and won’t build a roadmap from a placeholder. The agent never invents evidence — no evidence means inferred or unknown.

Behavior

• Requires kaddo init first.
• State-aware messages — no more “this project is not marked as new” warning on pre-ai/legacy.
• Never overwrites existing files — they are reported as skipped. An existing knowledge/knowledge.md, roadmap.md or work-items/ is kept.
• Idempotent — running it again writes nothing new.
• Does not install agents or skills — those stay kaddo add agents / kaddo add skills.

Where it fits

kaddo init → kaddo bootstrap → kaddo add agents → kaddo add skills → …

All Kaddo surfaces converge on kaddo bootstrap as the dominant recommendation when the baseline is incomplete:

• kaddo explain — suggests kaddo bootstrap first, then agents, skills, and context in order.
• kaddo context — suppresses agent handoff in both markdown and JSON output (recommendedAgents: [], “No agent handoff yet”), shows bootstrap-specific missing-context messages and bootstrap-aware LLM instructions.
• kaddo understand — shows “Agent handoff is not ready yet” with the bootstrap sequence (bootstrap → agents → skills → context → understand). No agent flow or prompt paths appear.
• kaddo route — includes a dedicated bootstrap step that is current until the baseline exists.
• MCP resources (kaddo://next-step, kaddo://project-route) — return the same bootstrap-first recommendation.

Next steps

kaddo add agents     # install agent prompt packs
kaddo add skills     # install reusable skills
kaddo scan           # (pre-ai/legacy) capture deterministic signals from the code
kaddo understand     # guided handoff

Kaddo prepares structure; your LLM and your team provide the content. Kaddo never invents business facts and never writes code.