Pilot semantic-tooling dogfooding during DEL-to-Workshop lowering

[Teakowa]

Related: #30
Parent integration track: #8

Goal

Use WrightKit's existing DEL/OSTW and Workshop semantic tooling as part of the real agent development workflow for core DEL -> Workshop lowering, and collect concrete evidence about where semantic tooling improves or obstructs implementation before expanding this workflow across WrightKit.

Context

del-rs already exposes Workshop-independent parse/check/HIR/inspect tooling, while workshop-rs owns canonical raw Workshop parsing, WIR, validation, and emission. #30 is the first production DEL HIR -> canonical WIR lowering task and therefore provides a useful real workload for testing whether coding agents can use WrightKit's own semantic capabilities to implement WrightKit more accurately.

This is dogfooding, not a new correctness oracle. Tool output from the implementation under development must remain advisory and must not replace independent corpus evidence, tests, canonical validation, or pinned external/reference evidence.

Scope

• During Lower core DEL HIR into canonical workshop-rs WIR #30 investigation and implementation, have coding agents preferentially use existing machine-readable del-rs tooling (check, hir, inspect, where applicable) to understand DEL source, semantic identities, provenance, and frontend behavior instead of relying only on textual search/manual source reading.
• Use workshop-rs parsing/WIR/validation tooling to inspect and validate the canonical Workshop side of representative lowering cases.
• Keep normal Rust tooling, focused tests, compatibility corpus evidence, and canonical WIR validation as independent validation layers.
• Record only concrete dogfooding findings encountered during real Lower core DEL HIR into canonical workshop-rs WIR #30 work, such as missing semantic queries, ambiguous or misleading diagnostics, insufficient provenance, unstable/unusable JSON output, excessive invocation cost, or cases where the tooling materially reduced investigation effort.
• Fix small, clearly owned tooling defects when doing so is necessary to continue Lower core DEL HIR into canonical workshop-rs WIR #30 and remains bounded; otherwise capture a focused follow-up issue in the owning repository rather than expanding Lower core DEL HIR into canonical workshop-rs WIR #30 scope.
• At the end of the pilot, summarize whether the workflow should expand to Implement DEL object, reference, dispatch, recursion, and closure runtime lowering #31/Integrate DEL project, embedded Workshop, settings, strings, and backend compiler surfaces #32, OPY integration, or Wright-wide agent development, and identify the smallest high-value tooling additions justified by observed usage.

Non-goals

• Making del-rs, Wright, or any candidate branch output the sole source of truth for its own correctness.
• Requiring a new LSP, MCP, agent protocol, telemetry system, or general plugin framework before the pilot can run.
• Adding speculative semantic-query APIs that were not needed by the real lowering work.
• Blocking Lower core DEL HIR into canonical workshop-rs WIR #30 on unrelated tooling imperfections when existing evidence and normal development tools remain sufficient.
• Applying the workflow as a mandatory WrightKit-wide policy before the pilot produces evidence.

Acceptance criteria

• At least representative Lower core DEL HIR into canonical workshop-rs WIR #30 lowering work is performed with active use of del-rs semantic tooling and workshop-rs canonical tooling, not only manual source inspection.
• Dogfooding never replaces independent tests/corpus/reference evidence or canonical Workshop validation.
• Material tooling failures or missing capabilities encountered during the work are captured with concrete examples and ownership rather than silently bypassed.
• The pilot records at least one evidence-backed conclusion about where semantic tooling helped, did not help, or misled the agent workflow.
• A closing recommendation states whether to expand the workflow to Implement DEL object, reference, dispatch, recursion, and closure runtime lowering #31/Integrate DEL project, embedded Workshop, settings, strings, and backend compiler surfaces #32 and later OPY/Wright development, and any proposed tooling additions are derived from observed needs rather than speculative API design.

Dependencies

• Lower core DEL HIR into canonical workshop-rs WIR #30 is the pilot workload.
• Establish the workshop-rs dependency and canonical catalog provider #29 supplies the canonical workshop-rs provider/dependency seam used by Lower core DEL HIR into canonical workshop-rs WIR #30.

Planning notes

Start using the available tools immediately. Prefer CLI/machine-readable semantic queries for agent workflows and treat LSP as an optional adapter rather than a prerequisite. Keep the pilot lightweight: the purpose is to expose real product friction quickly, not to build a self-hosting framework before evidence exists.