[Workflow] Can I use cavemen, ponytail kind of ai agent token usage helpers with workflow command?

[rocxteady]

Can I use cavemen, ponytail kind of ai agent token usage helpers with workflow command? Since workflow command creates and goes in agents cli system itself. I cant verify the ponytail is actually active when they are working. I feel like they dont work. How do i make my workflows work with other tools like this? Thanks.

[mnriem]

Why your token-usage monitors look inactive

When a workflow prompt or command step calls Copilot, it doesn't open an interactive session you can watch. It spawns a quick one-shot subprocess and exits — roughly:

copilot -p "…the prompt…" --agent speckit.<step> --yolo

A new copilot process is launched for each step, then it's gone. So a ccusage / ponytail-style helper running in your terminal never attaches to it — that's why it looks like nothing is happening.

The fix: wrap the Copilot binary

Spec Kit lets you swap in your own executable via an env var, so you can slip your usage tool into the chain:

# Tell Spec Kit's Copilot integration to launch this wrapper
# instead of the plain `copilot` binary.
export SPECKIT_INTEGRATION_COPILOT_EXECUTABLE=/usr/local/bin/copilot-with-usage

#!/usr/bin/env bash
# /usr/local/bin/copilot-with-usage
# Run the usage tracker, then hand off to the real copilot CLI
# with all the same arguments the workflow passed in ("$@").
exec ccusage wrap -- copilot "$@"

Now every step the workflow runs goes through your monitor, so it's guaranteed to be active.

Good to know

• Counts are per-step, not one long session — point your tool at Copilot's own usage logs if it groups by session.
• Need full control instead? Just call Copilot from a type: shell step: ccusage run -- copilot -p "...".

That's it — one env var plus a two-line wrapper, or a shell step.