diff --git a/packages/cloud-agents/src/server/workflows/__tests__/completionReportingSkill.test.ts b/packages/cloud-agents/src/server/workflows/__tests__/completionReportingSkill.test.ts
index e4b9999d7..469bd8604 100644
--- a/packages/cloud-agents/src/server/workflows/__tests__/completionReportingSkill.test.ts
+++ b/packages/cloud-agents/src/server/workflows/__tests__/completionReportingSkill.test.ts
@@ -20,4 +20,19 @@ describe('Standard completion-reporting skill guidance', () => {
'Mention checks only when one failed, was skipped or unavailable, materially changes confidence in the result, or the user explicitly asked for that detail, and when they do matter, mention them inline in the surrounding summary instead of as a dedicated status block.',
);
});
+
+ it('requires calibrated completion evidence without fixed report headings', () => {
+ const skillPath = path.resolve(
+ thisDirPath,
+ '../skills/standard/implement-changes/SKILL.md',
+ );
+ const skillContent = fs.readFileSync(skillPath, 'utf8');
+
+ expect(skillContent).toContain(
+ 'Make clear what was completed, what evidence supports it, the scope actually covered, and any material item not verified or still pending.',
+ );
+ expect(skillContent).toContain(
+ 'Keep those distinctions in natural prose rather than requiring fixed headings or boilerplate.',
+ );
+ });
});
diff --git a/packages/cloud-agents/src/server/workflows/skills/standard/implement-changes/SKILL.md b/packages/cloud-agents/src/server/workflows/skills/standard/implement-changes/SKILL.md
index 34ed2f0a6..4d6f90e23 100644
--- a/packages/cloud-agents/src/server/workflows/skills/standard/implement-changes/SKILL.md
+++ b/packages/cloud-agents/src/server/workflows/skills/standard/implement-changes/SKILL.md
@@ -175,6 +175,7 @@ You are a coding workflow specialist. Analyze the request, implement a correct r
The only valid terminal states for a repository-changing run are: delegated delivery completed, an explicit blocker, or an explicit policy pause awaiting user input.
Summarize behavior-level changes, not just file names.
Let the delegated delivery skill own pull request title/body derivation, screenshot and screencast embedding, related-PR links, and any PR metadata refresh using its shared `pr-metadata-update-recipe` block plus the relevant `pr-writing-guide` section instead of duplicating that procedure here.
+ Make clear what was completed, what evidence supports it, the scope actually covered, and any material item not verified or still pending. Keep those distinctions in natural prose rather than requiring fixed headings or boilerplate.
Keep the final completion report conversational and concise by default; do not turn routine successful execution into an audit log.
Do not add standalone `Validation`, `Checks`, or `Status` sections for routine successful runs.
Mention screenshot or screencast capture only when it materially affects the user's next step or the user explicitly asked about it.