[M12] Tooling-first lint and Workshop stability analysis platform

[Teakowa]

Goal

Make Wright's tooling-first product direction concrete by turning the existing Workshop analyses into a configurable lint/analysis platform and validating high-value server-stability rules against real projects.

M12 should deliver developer value before further broad source-language compatibility expansion.

Context

wright-analyzer already has a small hard-coded analysis set (min-wait-loop, duplicate-condition, expensive-loop-check) over WIR/CFG. The current analyze() path constructs a fixed array of analyses. That is sufficient evidence for a rule framework, but not for a large plugin architecture.

Workshop stability guidance is also not universally safe: an "anti-crash" rewrite or rule can itself create load or semantic regressions. Wright should therefore provide evidence-backed findings with explicit confidence/limits rather than encode community advice as unquestionable correctness rules.

ADR-0008 / #88 is the architecture authority. M11/#82 and #88 are complete, so M12 is now the active product-value milestone.

A post-#88 review identified useful future work around Workshop content data, extension boundaries, and independent version identities. That work is recorded in #96 and is explicitly non-blocking/deferred for M12 unless new implementation evidence proves otherwise.

Work items

• [M12] Establish lint rule registry and configuration contract #97 — Establish lint rule registry and configuration contract.
• [M12] Add first-class lint CLI and shared structured results #98 — Add first-class lint CLI and shared structured results.
• [M12] Ship a bounded evidence-backed Workshop stability rule set #99 — Ship a bounded evidence-backed Workshop stability rule set.
• [M12 acceptance] Verify tooling-first lint and stability platform #100 — Independently verify M12 acceptance.

Scope

• Define a Wright-owned lint/rule contract over WIR/CFG/semantic services with stable rule IDs, severity, finding metadata, source provenance, and documentation/known-limit fields.
• Move the current hard-coded analyses behind the rule registry without changing their observable findings unintentionally.
• Add deterministic configuration for enabling/disabling rules, severity overrides, and bounded rule options where actually required.
• Add a first-class wright lint workflow backed by the same compiler/session and structured-result contracts as check/analyze.
• Expose lint findings through existing embedding/tool/agent APIs rather than creating a separate semantic service.
• Build a small first-party Workshop stability/performance rule set chosen from real corpus evidence, including repeated/expensive evaluation and hot-loop/timing patterns where false-positive risk is acceptable.
• Reuse the existing cost/resource model where a rule makes resource claims; distinguish exact counts, static indicators, heuristics, and runtime-validated evidence.
• Preserve an extension-friendly core contract, but defer third-party rule loading/plugin-host implementation and broader extension/version architecture to [v0.2 workshop-rs] Define Workshop catalog, locale, extension, and version boundaries #96 until concrete evidence justifies it.

Non-goals

• Treating community "anti-crash" guidance as guaranteed-safe transformations.
• Claiming precise Overwatch server CPU/runtime cost without runtime evidence.
• Automatically applying risky performance rewrites.
• A large ESLint-compatible JavaScript runtime.
• Implementing or freezing a Rust/WASM/process plugin ABI in M12.
• Implementing [v0.2 workshop-rs] Define Workshop catalog, locale, extension, and version boundaries #96 Workshop content-pack/hot-update/versioning work.
• Expanding OPY/OSTW language syntax merely to increase lint coverage.
• Implementing OSTW support or Workshop -> source-language conversion in this milestone.

Deliverables

• Rule registry/configuration contract.
• wright lint CLI + machine-readable result integration.
• Existing analyses migrated to the registry.
• A bounded, corpus-validated first-party stability rule set.
• Agent/tool API exposure for lint queries/findings.
• Independent QA/Architect acceptance evidence.

Acceptance criteria

• The existing three analyzer findings run through the new rule framework and retain regression coverage.
• Users can enable/disable rules and override severity deterministically without source changes.
• wright lint works for supported OPY and raw Workshop inputs through the shared driver/session pipeline.
• Findings preserve original source identity/spans where available and expose structured rule IDs through CLI JSON and tool APIs.
• Every newly shipped stability rule has positive/negative fixtures plus at least one real-project evidence case or an explicit reason why only synthetic validation is possible.
• Heuristic findings are labeled as heuristics and do not masquerade as guaranteed runtime behavior.
• No rule or fixer is justified solely because a wiki/community source calls it "anti-crash".
• M12 does not require a third-party rule loader, content registry, or other [v0.2 workshop-rs] Define Workshop catalog, locale, extension, and version boundaries #96 implementation to ship.
• [M12 acceptance] Verify tooling-first lint and stability platform #100 independently verifies the current implementation and CI before M12 closes.
• PM records a post-M12 reassessment before decomposing [M13] Source-language interoperability and OSTW compatibility #90/M13.

Planning notes

Keep implementation focused on #97 -> #98 -> #99 -> #100. #97 may begin immediately; #98 follows the registry contract; #99 should use the proven registry/tooling surface rather than designing around hypothetical plugins. Do not mechanically expand OPY compatibility or begin OSTW work during M12 unless a concrete tooling blocker is demonstrated.

[Teakowa]

The architecture rebaseline in #88 (commit c96fd08) is now recorded. M12 depends on it per the issue scope. No prioritization change is needed for M12; the lint platform work is the primary vehicle that makes the tooling-first direction concrete.