docs: describe the java-functional-style companion boundary

AGENTS.md

@@ -19,6 +19,7 @@ that correction durable in the relevant file under `docs/agents/`.
 - [Project Identity](docs/agents/project-identity.md)
 - [README Guidance](docs/agents/readme.md)
 - [Skill Behavior](docs/agents/skill-behavior.md)
+- [Ownership Boundaries](docs/agents/ownership-boundaries.md)
 - [Eval Guidance](docs/agents/evals.md)
 - [Public Metadata And OSS Readiness](docs/agents/public-metadata.md)
 - [Maintaining Agent Docs](docs/agents/maintaining-agent-docs.md)

README.md

@@ -18,12 +18,16 @@ as a design choice rather than a default optimization.
 It also tells the agent to check the project Java version first. The right stream code for Java 8
 may be different from the right code for Java 17, Java 21, or Java 24.
 
+General lambda and callback style lives in the companion package
+`martinfrancois/java-functional-style`; see [Ownership Boundaries](#ownership-boundaries).
+
 ## Contents
 
 - [Getting Started](#getting-started)
 - [Why This Exists](#why-this-exists)
 - [Common Stream Mistakes](#common-stream-mistakes)
 - [What It Helps With](#what-it-helps-with)
+- [Ownership Boundaries](#ownership-boundaries)
 - [How It's Evaluated](#how-its-evaluated)
 - [Origin](#origin)
 - [Contributing](#contributing)
@@ -241,6 +245,23 @@ Poor fit:
 - large API redesigns or new dependencies without maintainer agreement;
 - changing business behavior just to make code look more functional.
 
+## Ownership Boundaries
+
+`martinfrancois/java-streams` owns stream and collector semantics. General lambda and callback
+style (identity functions, no-op stages, helper extraction, comparator composition,
+method-reference pitfalls, supplier laziness, checked boundaries in callbacks) is owned by the
+companion package `martinfrancois/java-functional-style`
+([repository](https://github.com/martinfrancois/java-functional-style-skill)), and Optional
+semantics by `martinfrancois/java-optionals`
+([repository](https://github.com/martinfrancois/java-optionals-skill)).
+
+Each package works on its own. Install the companion next to this one when stream cleanup also
+involves non-trivial callbacks:
+
+```bash
+npx tessl i martinfrancois/java-functional-style
+```
+
 ## How It's Evaluated
 
 The skill is tested on Java stream implementation, review, and cleanup tasks. Each task is run

docs/agents/ownership-boundaries.md

@@ -0,0 +1,47 @@
+# Ownership Boundaries
+
+## Scope
+
+Use this when changing runtime guidance, README wording, evals, or companion-package documentation
+for Java Streams. This page holds the only full list of who owns what; other pages link here.
+
+## Rules
+
+- `java-streams` owns stream and collector semantics: terminal operation choice, collector choice,
+  duplicate-key and null behavior, encounter order, primitive streams, `findFirst` versus
+  `findAny`, `parallelStream`, `Gatherers.mapConcurrent`, stream Java-version compatibility, and
+  stream-specific behavior preservation. It also keeps its own short "lambdas as glue" rule and its
+  preference for method references, because the published stream evals measure them.
+- `martinfrancois/java-functional-style` owns general Java lambda and functional-interface style:
+  identity functions, no-op functional stages, helper extraction from block callbacks, comparator
+  composition, method-reference pitfalls (receiver binding, overloads, boxing), supplier laziness,
+  checked-exception boundaries inside callbacks, and callback side-effect boundaries.
+- `martinfrancois/java-optionals` owns Optional semantics.
+- Each package works on its own. Don't make stream guidance depend on the companion being installed,
+  and don't remove stream guidance because the companion also covers it.
+- Don't grow generic lambda guidance here. When a functional-style gap shows up in stream work, fix
+  it in the companion and add stream-side text only if hosted evidence proves the stream evals need
+  it.
+
+## Composition Check
+
+Adding the companion must not make this skill worse. The check runs this repository's evals,
+unchanged, with both skills injected as context and requires 100% with-context for every scenario
+in the run:
+
+```bash
+# from a checkout of java-functional-style-skill, next to this repository
+scripts/run_composed_eval.sh ../java-streams-skill main
+```
+
+The main suite is required whenever either package changes runtime text; run `reference` and
+`regression` too when the change touches review wording or when budget allows. The companion
+repository owns the runner and runs the check before its releases; this repository runs it when
+its own runtime changes. Local validation alone doesn't satisfy this check.
+
+## References
+
+- [Skill Behavior](skill-behavior.md)
+- [README Guidance](readme.md)
+- [Eval Guidance](evals.md)
+- [Workflow](workflow.md)

docs/agents/skill-behavior.md

@@ -42,6 +42,8 @@ guidance, or auto-selection wording.
 - Keep stream lambdas as short glue. Prefer method references or one-expression lambdas whose body
   stays on the same line as `->`, and extract named helpers for branching, loops, temporary
   variables, formatting, merge rules, or nested stream chains that would continue on later lines.
+  This rule stays here because the published stream evals measure it; deeper callback-style
+  guidance belongs to the companion package, see [Ownership Boundaries](ownership-boundaries.md).
 - Runtime guidance should keep internal workflow language out of ordinary user-facing reviews. Avoid
   terms such as "hard stop", "marker", "scan", "checklist", and skill names unless the user asked
   for an explicit skill workflow, audit, or scan command.
@@ -52,3 +54,4 @@ guidance, or auto-selection wording.
 
 - [README Guidance](readme.md)
 - [Eval Guidance](evals.md)
+- [Ownership Boundaries](ownership-boundaries.md)

docs/agents/workflow.md

@@ -126,6 +126,9 @@ release-readiness.
 
   Follow the recommendation unless the pull request documents a maintainer-approved override.
 
+- When runtime skill text changes, also run the composition check in
+  [Ownership Boundaries](ownership-boundaries.md) so the companion package still composes cleanly.
+
 - Run the Tessl skill quality review at threshold 100 when changing runtime skill content:
 
   ```bash