Spike executable contract projection on one command family

[devin-ai-integration]

Question

Can one executable, canonical command-family contract reduce duplicated client/MCP projection work
without adding more machinery than it removes?

Time-boxed spike

Choose one representative already-typed command family and prototype:

• canonical input/output contracts in a neutral layer;
• client method/result types projected from those contracts;
• MCP output schema produced from an executable contract or a tightly colocated schema;
• operational traits remaining on CommandDescriptor;
• parity tests proving unchanged CLI, daemon, client, and MCP behavior.

Measure the before/after number of independently edited projection declarations and the runtime,
bundle, dependency, and maintenance cost of the mechanism.

Stop conditions

Stop and document the result rather than generalizing if the spike requires:

• generated source or a new schema DSL/dependency;
• cold-start or bundle regression;
• new import back-edges;
• a parallel registry;
• weaker completeness/parity checks;
• more hand-authored declarations than the current design.

Deliverable

A small experimental PR or written design result with measured evidence and an explicit
continue/stop recommendation. Do not begin a repository-wide migration in this issue.

Blocked by: none technically, but do not combine with the public client-result narrowing track.

Origin: post-#1174 architecture follow-up.

[devin-ai-integration]

Completed by #1208 and finalized in #1213: the selected navigation family reduced projection declarations 15 → 5 with measured bundle/runtime evidence and explicit stop conditions. No repository-wide migration is implied.