Query tool_calls table for the 5 skill_ids with the highest failure rate (status='failed') in the last 14 days. For each failing tool: (1) read the error_message patterns across all failures; (2) identify root cause (API change, schema mismatch, timeout, auth); (3) implement a fix or workaround in the tool code; (4) write a test call to verify the fix works; (5) update the skills table with fix notes. Document all findings in a work log. Restoring failed tools directly increases the capabilities of all agents.
Completion Notes
Auto-completed by supervisor after successful deploy to main