Wire test, docs, and report conventions into builder chain

plugins/corbits-skills/skills/implement/SKILL.md

@@ -92,7 +92,9 @@ The order of operations depends on whether you're fixing a bug or building a fea
 
 Keep the test focused on the behavior introduced by this commit. Don't test unrelated functionality. The test is part of the deliverable, not an afterthought.
 
-Keep the scope tight to what was discussed. If you discover additional work is needed, finish the current commit's scope first and note the additional work for a future commit.
+The test lands in the same unit of work as the implementation (Step 4) — same commit when committing — one logical unit (source of truth: style skill, AGENTS.md). When the caller passes testsmith-designed cases, land them as the implementation tests; any case left unlanded goes in the report with why so the caller can route a tester run.
+
+Keep the scope tight to what was discussed. If you discover additional work is needed, finish the current commit's scope first and note the additional work for a future commit. When the landing alters documented behavior beyond the discussed doc scope, flag it for the caller so a shakespeare docs pass can follow.
 
 ### Step 3: Build Gate
 
@@ -105,12 +107,13 @@ Run `make` (or the project's equivalent full pipeline: format, lint, build, test
 - If the build fails due to pre-existing issues unrelated to your changes, report the failure to the caller and let them decide how to proceed
 - Do not move forward with a broken build
 - Do not substitute partial builds (e.g., running only the compiler) for the full pipeline
+- Record the exact verification commands and their exit statuses: the report maps each success criterion to pass, fail, or blocked with command evidence
 
 ### Step 4: Commit
 
 Update `activeForm` to "Committing: {subject}".
 
-Create the commit. Follow the commit message conventions from the `style` skill. Include the test in the same commit as the implementation — they are one logical unit of work.
+Create the commit. Follow the commit message conventions from the `style` skill. Include the test in the same unit of work as the implementation — same commit when committing — one logical unit — and update the docs when the commit changes documented behavior. Worker-chain branch/PR convention: branch name carries the issue id, the PR body ends with `Fixes CL-…` and carries no AI-attribution lines (CONTRIBUTING: title stays a plain-English sentence, body is Summary/Verification).
 
 ### Step 5: Critique Loop
 

src/agent/directors/builder/package.test.ts

@@ -167,6 +167,30 @@ describe("builderPackage", () => {
     expect(prompt).toMatch(/Summary \/ Findings \/ Blockers \/ Paths/);
   });
 
+  test("systemPrompt wires same-unit tests, docs upkeep, and report mapping", () => {
+    const p = builderPackage.systemPrompt;
+    expect(p).toMatch(/same commit/);
+    expect(p).toMatch(/same commit when committing/);
+    expect(p).toMatch(/alters documented behavior/i);
+    expect(p).toMatch(/update the docs/i);
+    expect(p).toMatch(
+      /map each success_criteria item to pass, fail, or blocked/,
+    );
+    expect(p).toMatch(
+      /bare .*pass.*without command evidence.*incomplete report/is,
+    );
+  });
+
+  test("systemPrompt wires docs routing, testsmith consumer, and branch/PR shape", () => {
+    const p = builderPackage.systemPrompt;
+    expect(p).toMatch(/testsmith-designed cases/);
+    expect(p).toMatch(/route a tester run/);
+    expect(p).toMatch(/shakespeare docs pass/);
+    expect(p).toMatch(/branch name carries the issue id/i);
+    expect(p).toMatch(/Fixes CL-/);
+    expect(p).toMatch(/no AI-attribution lines/);
+  });
+
   test("systemPrompt preserves public API sync/async under Guidelines", () => {
     const prompt = builderPackage.systemPrompt;
     expect(prompt).toMatch(/Public API shapes/i);

src/agent/directors/builder/package.ts

@@ -60,6 +60,8 @@ The order of operations depends on whether you're fixing a bug or building a fea
 
 Keep the test focused on the behavior introduced by this unit of work. Don't test unrelated functionality. The test is part of the deliverable, not an afterthought.
 
+Land the test in the same unit of work as the implementation — same commit when committing — one logical unit. When the change alters documented behavior, update the docs that describe it in the same unit of work (source of truth: style skill, AGENTS.md). When the brief carries testsmith-designed cases, land them as the implementation tests; any case left unlanded goes under Blockers with why so the parent can route a tester run. When the landing alters documented behavior outside the brief's doc scope, flag it under Blockers so the parent can route a shakespeare docs pass.
+
 Keep the scope tight to the brief. If you discover additional work is needed, finish the current brief's scope first and note the additional work under Blockers / Findings for a future unit.
 
 ## Build Gate
@@ -79,7 +81,7 @@ For implementation work, run the repository-defined typecheck command and releva
 
 **Don't shortcut verify.** The value is in the discipline. Skipping the build gate "because this change is simple" defeats the purpose.
 
-**Keep units focused.** Deliver a working tree that satisfies the brief and report. Builder does NOT commit unless the brief's success_criteria explicitly ask for a commit — the parent / Skywalker usually owns commits. Prefer: working tree + report envelope.
+**Keep units focused.** Deliver a working tree that satisfies the brief and report. Builder does NOT commit unless the brief's success_criteria explicitly ask for a commit — the parent / Skywalker usually owns commits. Prefer: working tree + report envelope. Worker-chain branch/PR convention for the parent's handoff: branch name carries the issue id, the PR body ends with \`Fixes CL-…\` and carries no AI-attribution lines (CONTRIBUTING: title stays a plain-English sentence, body is Summary/Verification only).
 
 **Discovered extra work** belongs under Blockers / Findings for a future unit — finish the current brief first.