Claude Tag Observability in Scorecard
TLDR: Trace Claude Tag in your org - https://docs.scorecard.io/intro/claude-tag-tracing
Claude Tag is Anthropic’s newest integration with Slack, allowing Claude to take action and respond to messages in your private channels.
At Scorecard, we’ve integrated tracing natively into Claude Tag, enabling a self improvement loop on our newest coworker.
Claude Tag uses a remote Claude Code session under the hood, allowing for the same 0-code Scorecard tracing integration.
In Scorecard, you can see Claude Tag’s:
- MCP tool calls
- Subagents and their traces
- Session context and memory used
- Reasoning for deciding whether to reply to you
At Scorecard, we’ve been using these traces to identify failure modes in Claude Tag:
- Using memories instead of skills
- Losing context across threads
- Lack of self verification of actions taken
- Not storing feedback in memory
- Tool calls and API services failing
Understanding Tag’s failure modes has allowed us to fix our system prompts, skills, and APIs.
This feature is live for anyone to integrate in under 5 minutes. Create a new Claude Tag environment with your Scorecard API key, add the environment to your Slack permissions, and watch new Tag traces stream into Scorecard. Learn more in our docs: https://docs.scorecard.io/intro/claude-tag-tracing.
