From d38917f651a15dc2148a7a3febb12a5bf0037944 Mon Sep 17 00:00:00 2001
From: "github-actions[bot]"
 <41898282+github-actions[bot]@users.noreply.github.com>
Date: Fri, 4 Sep 2026 21:08:30 +0000
Subject: [PATCH] Replace unactionable code-improvement tip with scoping
 guidance
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The code-improvement archetype tip referenced internal scan-data
workflow names (pr-fix, ci-doctor) that aren't exposed anywhere in the
wizard UI or generator. Downstream agents building a new workflow from
this generated prompt have no way to act on 'avoid these templates' —
it's noise, not guidance. Replaced with the underlying, actionable
lesson: scope each code-improvement run to one narrow, well-defined
fix rather than an open-ended mandate, which is what the low success
rate of those degraded workflows actually reflects.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
---
 patterns/archetypes/code-improvement.json | 2 +-
 scripts/generate-patterns.py              | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/patterns/archetypes/code-improvement.json b/patterns/archetypes/code-improvement.json
index 0ab8236..3356f40 100644
--- a/patterns/archetypes/code-improvement.json
+++ b/patterns/archetypes/code-improvement.json
@@ -55,7 +55,7 @@
   "tips": [
     "Use schedule triggers for continuous maintenance coverage",
     "Add pre-steps to run tests/linters before the agent starts \u2014 validates baseline",
-    "Avoid pr-fix and ci-doctor templates \u2014 both have <20% success in practice",
+    "Scope each run to one narrow, well-defined improvement (a single failing check or a named category of fix) rather than an open-ended \"fix anything broken\" mandate \u2014 broad, unscoped code-improvement workflows have the lowest observed success rates",
     "Use explicit DO NOT constraints to keep changes scoped to the requested improvement",
     "Use protected-files: fallback-to-issue for manifests, CI configuration, and agent instructions",
     "Add a pre-activation step that skips scheduled runs once too many open PRs share your title prefix, to avoid spamming maintainers"
diff --git a/scripts/generate-patterns.py b/scripts/generate-patterns.py
index 344d286..e6a314b 100644
--- a/scripts/generate-patterns.py
+++ b/scripts/generate-patterns.py
@@ -204,7 +204,7 @@ def classify(wf):
             "tips": [
                 "Use schedule triggers for continuous maintenance coverage",
                 "Add pre-steps to run tests/linters before the agent starts — validates baseline",
-                "Avoid pr-fix and ci-doctor templates — both have <20% success in practice",
+                "Scope each run to one narrow, well-defined improvement (a single failing check or a named category of fix) rather than an open-ended \"fix anything broken\" mandate — broad, unscoped code-improvement workflows have the lowest observed success rates",
                 "Use explicit DO NOT constraints to keep changes scoped to the requested improvement",
                 "Use protected-files: fallback-to-issue for manifests, CI configuration, and agent instructions",
                 "Add a pre-activation step that skips scheduled runs once too many open PRs share your title prefix, to avoid spamming maintainers",