ci: optimize link checker workflow and failure reporting (#2411)

The current system streamlines troubleshooting by omitting passing links
from the output. Reporting now focuses solely on:

- Broken or failing links
- Redirecting links
- File paths (now prioritized at the top of the log for rapid
identification)

The link resolution notice is now presented first, followed by specific
errors to prevent developer distraction.

The .lycheeignore file has been updated to use regular expressions for
broader coverage of npm and SQL links, replacing the previous method of
listing individual URLs.

---------

Co-authored-by: Twisha Bansal <58483338+twishabansal@users.noreply.github.com>

.github/workflows/link_checker_workflow.yaml

@@ -32,29 +32,37 @@ jobs:
           restore-keys: cache-lychee-
 
       - name: Link Checker
+        id: lychee-check       
         uses: lycheeverse/lychee-action@a8c4c7cb88f0c7386610c35eb25108e448569cb0 # v2
+        continue-on-error: true 
         with:
           args: >
-            --verbose
+            --quiet
             --no-progress
             --cache
             --max-cache-age 1d
             --exclude '^neo4j\+.*' --exclude '^bolt://.*'
             README.md
             docs/
-          output: /tmp/foo.txt
-          fail: true
-          jobSummary: true
-          debug: true
+          output: lychee-report.md
+          format: markdown
+          fail: true           
+          jobSummary: false    
+          debug: false
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
-      # This step only runs if the 'lychee_check' step fails, ensuring the 
-      # context note only appears when the developer needs to troubleshoot.
-      - name: Display Link Context Note on Failure
-        if: ${{ failure() }}
+
+      - name: Display Failure Report
+        # Run this ONLY if the link checker failed
+        if: steps.lychee-check.outcome == 'failure'
         run: |
           echo "## Link Resolution Note" >> $GITHUB_STEP_SUMMARY
           echo "Local links and directory changes work differently on GitHub than on the docsite." >> $GITHUB_STEP_SUMMARY
           echo "You must ensure fixes pass the **GitHub check** and also work with **\`hugo server\`**." >> $GITHUB_STEP_SUMMARY
+          echo "See [Link Checking and Fixing with Lychee](https://github.com/googleapis/genai-toolbox/blob/main/DEVELOPER.md#link-checking-and-fixing-with-lychee) for more details." >> $GITHUB_STEP_SUMMARY
           echo "---" >> $GITHUB_STEP_SUMMARY
 
+          echo "### Broken Links Found" >> $GITHUB_STEP_SUMMARY
+          cat ./lychee-report.md >> $GITHUB_STEP_SUMMARY
+
+          exit 1

.lycheeignore

@@ -23,8 +23,7 @@ https://cloud.dgraph.io/login
 https://dgraph.io/docs
 
 # MySQL Community downloads and main site (often protected by bot mitigation)
-https://dev.mysql.com/downloads/installer/
-https://www.mysql.com/
+^https?://(.*\.)?mysql\.com/.*
 
 # Claude desktop download link
 https://claude.ai/download
@@ -37,10 +36,9 @@ https://dev.mysql.com/doc/refman/8.4/en/sql-prepared-statements.html
 https://dev.mysql.com/doc/refman/8.4/en/user-names.html
 
 # npmjs links can occasionally trigger rate limiting during high-frequency CI builds
-https://www.npmjs.com/package/@toolbox-sdk/adk
-https://www.npmjs.com/package/@toolbox-sdk/core
-https://www.npmjs.com/package/@toolbox-sdk/server
+^https?://(www\.)?npmjs\.com/.*
+
 https://www.oceanbase.com/
 
 # Ignore social media and blog profiles to reduce external request overhead
-https://medium.com/@mcp_toolbox
+https://medium.com/@mcp_toolbox