Kotlin Spring Review Workflow

Run all four mandatory skills against one immutable source snapshot and return one consolidated review. Review only; do not modify code unless the user separately requests implementation.

Exact dependencies

• kotlin-spring-review
• oop-design-review
• design-pattern-apply
• refactoring-catalog

Treat every dependency as mandatory. Never invoke kotlin-spring-review-workflow recursively or replace a missing dependency with a similarly named skill.

Cost-aware native routing

Resolve portable profiles from the current runtime capability catalog immediately before dispatch. Use kotlin-spring-review=standard/medium and oop-design-review|design-pattern-apply|refactoring-catalog=frugal/low by default. Raise only the affected pass by one tier when a verifier rejects its evidence or a confirmed transaction, security, irreversible-data, or public-contract risk exceeds the selected profile.

Every native handoff uses model_selection: explicit-native-override; spawn with fork_turns: none, the exact resolved model, and exact reasoning_effort. Never inherit the primary model or reasoning setting. Record requested and runtime-reported models separately; mismatch makes the run INCOMPLETE. Timeout, quota, authentication, and missing tools are not capability escalation signals.

Input contract

Resolve and freeze:

• repository root and exact file, module, diff, commit, or pull-request range
• comparison baseline, or none for a snapshot review
• applicable repository instructions and compatibility constraints
• build and test commands that may be run

Record the revision and file list before dispatch. Give every specialist the same scope and source snapshot. If the revision changes, mark every result STALE and return INCOMPLETE.

Workflow

1. Preflight

Verify all four skills by exact frontmatter ID. Capture changed symbols, affected callers, transaction boundaries, persistence mappings, build system, and existing tests. Stop before review with MISSING_DEPENDENCY or UNSTABLE_SOURCE when the mandatory suite cannot run consistently.

2. Parallel specialist pass

Use one bounded native background child per skill when the host exposes native subagents. Start at most four concurrently and forbid recursive delegation. If native children are unavailable, run the same four passes sequentially in the primary context; never silently drop a pass.

Require each result to include severity, confidence, path:line, affected symbol, root cause, evidence, impact, recommendation, governing principle, and verification.

• kotlin-spring-review: Kotlin idioms, nullability, coroutines, Spring DI, transactions, JPA, and framework correctness
• oop-design-review: responsibility placement, encapsulation, cohesion/coupling, dependency direction, and persistence leakage
• design-pattern-apply: real change axes, pattern fit, simpler alternatives, and over-design cost
• refactoring-catalog: code smells, behavior-preserving transformation sequence, and regression protection

Return only structured findings and artifact references to the primary context. Do not merge raw child transcripts.

3. Reconcile

Deduplicate only findings with the same location or symbol and the same root cause. Preserve all contributors and the strongest supported severity. Keep separate findings when one location has different causes.

Resolve common conflicts explicitly:

• pattern introduction versus KISS/YAGNI
• moving behavior into an entity versus JPA and transaction invariants
• Kotlin concision versus readability or framework proxy constraints
• refactoring cleanliness versus public behavior and compatibility

Prefer the smallest behavior-preserving action supported by evidence. Preserve unresolved dissent, including each skill's original severity, recommendation, and evidence. Do not invent consensus.

4. Verdict

• INCOMPLETE: any mandatory pass failed, timed out, was stale, or returned unusable output
• BLOCK: at least one supported CRITICAL finding
• CHANGES_REQUIRED: no CRITICAL finding and at least one supported HIGH finding
• WARNING: only supported MEDIUM or LOW findings remain
• APPROVE: all four passes completed and no supported finding remains

Never approve a partial run. Treat inferred CRITICAL findings as conditional until the named verification confirms their factual premise.

Output contract

Return one report containing:

1. frozen scope, baseline, constraints, assumptions, and commands run
2. four-entry run ledger with exact skill ID, status, and artifact reference
3. aggregate verdict and one-sentence rationale
4. unified findings with stable ID, location, evidence, contributors, and dissent
5. accepted, rejected, and unresolved recommendations with rationale
6. minimal before/after Kotlin examples for supported changes
7. compile, test, transaction, persistence, and regression verification matrix
8. coverage gaps and next actions

Redact credentials and personal data. A specialist's completion claim is candidate evidence; verify locations and build/test evidence before using it in the verdict.