⚠️ Potential issue | 🟡 Minor

# First, let's examine the actual code at the specified lines
cat -n internal/steps/response_builder.go | sed -n '140,150p'

Incomplete URL encoding for Shields.io badges—underscores not escaped.

Labels containing underscores will render incorrectly. Per Shields.io escaping rules, literal underscores must be escaped as __ in the badge URL. The current code handles spaces and dashes correctly but misses underscore escaping. Additionally, if labels contain reserved characters (%, #, ?, &, /, +), they should be percent-encoded.

 	// URL encode label
 	encodedLabel := strings.ReplaceAll(label, " ", "%20")
+	encodedLabel = strings.ReplaceAll(encodedLabel, "_", "__")
 	encodedLabel = strings.ReplaceAll(encodedLabel, "-", "--") // Shields.io escaping
 	badges = append(badges, fmt.Sprintf("![](https://img.shields.io/badge/%s-%s)", encodedLabel, color))

In `@internal/steps/response_builder.go` around lines 144 - 147, The badge label
encoding currently only replaces spaces and dashes (see encodedLabel handling)
but misses escaping underscores and other reserved URL characters; update the
encoding for encodedLabel (used before appending to badges in the badges
construction) to: first escape literal underscores by replacing "_" with "__"
per Shields.io rules, then percent-encode remaining reserved characters (use
net/url.QueryEscape or equivalent) so characters like %, #, ?, &, /, + are
encoded; ensure the final string is what you insert into
fmt.Sprintf("![](https://img.shields.io/badge/%s-%s)", encodedLabel, color).

This session file appears to be unrelated to the current PR's purpose. The session references branch "feature/hybrid-index-learn" but this PR is about message template redesign. This file seems to have been accidentally included in this pull request and should be removed or committed separately in the appropriate branch/PR.

The label encoding logic only handles spaces and hyphens, but labels can contain other special characters that need URL encoding (e.g., slashes, underscores, parentheses). Additionally, the current logic first replaces spaces with %20, then replaces hyphens with --. However, if a label already contains "%20" from a previous step or other URL-encoded characters, the hyphen replacement could affect those. Consider using a more comprehensive URL encoding function like url.QueryEscape() from the net/url package, or apply the Shields.io escaping rules more carefully (hyphens should be escaped to -- before other encoding).

The title truncation logic has a potential issue with multibyte UTF-8 characters. Using title[:47] could slice in the middle of a multibyte character, resulting in invalid UTF-8 and potential display issues. Consider using the utf8 package or []rune conversion to safely truncate at character boundaries: runes := []rune(title); if len(runes) > 50 { title = string(runes[:47]) + "..." }.

The color assignment heuristic using string matching (lines 141-142) is fragile and may not work well with all label types. Consider making the color mapping configurable or using label metadata if available. For example, labels might have a color property from GitHub that could be used, or the configuration could include a mapping of label patterns to badge colors. This would make the system more flexible and maintainable.

The message "To check" appears to be grammatically incorrect or unclear. When the issue belongs in the current repository, the message should probably say "Belongs in" or "Correctly placed in" instead of "To check". The phrase "To check" suggests an action that needs to be taken, which doesn't match the semantic meaning of confirming the issue is in the right place.

The test expectation doesn't match the implementation. The test expects "To check" but this wording is unclear. The test should verify the actual message that will be displayed, which should clearly indicate the issue belongs in the current repository (e.g., "Belongs in" or "Correctly placed in").

⚠️ Potential issue | 🟡 Minor

head -n 260 internal/steps/response_builder.go | tail -n 30

cat -n internal/steps/response_builder.go | sed -n '240,255p'

cat -n internal/steps/response_builder.go | sed -n '200,280p'

head -n 30 internal/steps/response_builder.go

rg -n "append.*parts.*fmt\.Sprintf" internal/

rg -n "Markdown|markdown|table|Table" internal/steps/response_builder.go

| Expr | Meaning |
|---|---|
| a \| b | OR |
| a | b | OR |

# Check if there are other places in the file building tables or formatted Markdown with user data
rg -n "similar\.|Reasoning|Title|State" internal/steps/response_builder.go | head -20

Escape table-sensitive characters in similar issue titles.

The title is embedded in both a Markdown table cell and link text. Raw titles containing |, [, ], or newlines will break table parsing. For example, a title like "foo | bar" renders as | ... | [#123 foo | bar](...) | ..., breaking the column structure.

Sanitize by replacing newlines/carriage returns with spaces, then escaping |, [, and ] with backslashes using strings.NewReplacer().

-		title := similar.Title
+		title := strings.ReplaceAll(similar.Title, "\n", " ")
+		title = strings.ReplaceAll(title, "\r", " ")
 		if len(title) > 50 {
 			title = title[:47] + "..."
 		}
+		title = strings.NewReplacer("|", "\\|", "[", "\\[", "]", "\\]").Replace(title)

 		parts = append(parts, fmt.Sprintf("| %.0f%% | [#%d %s](%s) | %s |",
 			similar.Similarity*100, similar.Number, title, similar.URL, status))

In `@internal/steps/response_builder.go` around lines 243 - 250, The similar issue
title isn't escaped before embedding in the Markdown table, so sanitize
similar.Title by first replacing newlines/carriage returns with spaces and then
escaping table-sensitive characters (|, [, ]) using strings.NewReplacer, assign
the escaped value to title (or titleEscaped) and perform truncation on the
escaped string (or ensure truncation preserves escapes), then use that
escaped/truncated title in the fmt.Sprintf that appends to parts (referencing
the variables similar.Title, title, parts and the fmt.Sprintf call).