Description
The description frontmatter in agents/coach.md contains four <example> blocks. Each example shows the assistant explicitly saying it will "launch the pair-programmer agent via the Task tool":
assistant: "I'll use the Task tool to launch the pair-programmer agent since you've declared Level 2 assistance."
Two problems:
- Outdated invocation language — The current Claude Code subagent framework delegates naturally without the assistant narrating tool calls. Examples should model the actual delegation behaviour, not the old explicit Task-tool pattern.
- Wrong agent ID — Two of the four examples refer to
the pair-programmer agent; one correctly uses pair-programmer:coach. The description should consistently use the canonical agent ID pair-programmer:coach.
Impact
Medium — incorrect examples in the description field can degrade auto-invocation accuracy since the description is used by the main Claude instance to decide when and how to delegate.
Proposed Fix
Rewrite the <example> blocks to reflect current delegation patterns (no explicit "Task tool" narration) and use the consistent agent ID pair-programmer:coach throughout.
Description
The
descriptionfrontmatter inagents/coach.mdcontains four<example>blocks. Each example shows the assistant explicitly saying it will "launch the pair-programmer agent via the Task tool":Two problems:
the pair-programmer agent; one correctly usespair-programmer:coach. The description should consistently use the canonical agent IDpair-programmer:coach.Impact
Medium — incorrect examples in the
descriptionfield can degrade auto-invocation accuracy since the description is used by the main Claude instance to decide when and how to delegate.Proposed Fix
Rewrite the
<example>blocks to reflect current delegation patterns (no explicit "Task tool" narration) and use the consistent agent IDpair-programmer:coachthroughout.