Design-time capabilities

The design-time tools turn a plain-language request into Chronicle artifacts — read models, projections, audits, and catalogs — grounded in the store’s real schema. They are read-only against the store and generative against your codebase: nothing is written to the store, and generated code comes back as a proposal you review. See How it works for the principle behind them.

Capabilities

Capability	Tool	Use it to
Describe the system	describe_system	Deduce what the system is and is for — entities, lifecycles, read surfaces, automations — and tell its story.
Suggest the next event types	suggest_next_event_types	Find lifecycle gaps and get grounded suggestions for the event types to introduce next.
Ad-hoc projection	run_ad_hoc_projection	Answer “show all X with all details” by folding events into current state on demand — nothing registered.
Describe an event type	describe_event_type	Read an event type’s real fields and types — the grounding primitive the others build on.
Read-model scaffolding	scaffold_read_model	Turn a set of event types into a reviewable read model + projection, grounded in their schema.
Unconsumed-event audit	audit_unconsumed_event_types	Find events nothing reads, and consumers pointing at event types that no longer exist.
Event catalog	generate_event_catalog	Produce a living data dictionary — every event, its fields, and its consumers.
Causal trace	explain_causal_trace	Turn an event source’s raw log into a “what happened and why” narrative.

Prompts

Beyond tools, the server exposes MCP prompts that package whole workflows — clients that surface prompts (for example as slash commands) can offer them directly:

Prompt	What it does
describe_system	Describe what the system is and is for, and tell its story.
suggest_next_event_types	Propose the next event types, refined with domain knowledge.
query_system	Answer a natural-language question (“show all employees with all details”) with an ad-hoc projection.

How an agent combines them

A typical flow chains a few tools. To answer “Show all employees with all details” an agent might:

1. Understand the system with describe the system — which entities exist and which event types carry which facts.
2. Fold the relevant event types into current state with an ad-hoc projection and present the result.
3. If the question keeps coming back, propose a permanent read model with read-model scaffolding.

Because the tools return structured data rather than prose, the agent does the natural-language reasoning while the store supplies the facts.

Roadmap

The capability hand-off sketches ten design-time capabilities. The tools above cover the grounding-and-generation core plus system understanding, evolution suggestions, and ad-hoc querying. The remaining ideas — projection-vs-reducer advice, consistency-boundary (DCB) advice, schema-evolution assistance, spec scaffolding, and constraint suggestions — build on the same introspection pipeline and are candidates for future releases.