The model-facing prompts behind Claude Code's built-in slash commands — the ones a user
invokes by typing /name. These are distinct from bundled skills (see ../bundled-skills/):
they register as type:"prompt" or type:"local" commands, not as skills.
| File | Command | What it does |
|---|---|---|
btw.md |
/btw |
spawns a lightweight side-question agent that answers without interrupting the main thread |
compact.md |
/compact |
summarizes the conversation to free context |
compact-continuation-message.md |
— | the message injected into the continued session after a compaction |
compact-rewind-summarization.md |
— | the summarization prompt used for rewind/auto-compaction |
recap.md |
/recap |
the one-line "you stepped away" recap (shared with the auto away-summary) |
insights.md |
/insights |
generates a session-analysis report |
team-onboarding.md |
/team-onboarding |
builds a teammate onboarding guide from usage data |
session-title.md |
/rename |
auto-generates a session title (also used for automatic naming) |
Captured verbatim from the Claude Code 2.1.211 binary. Runtime data slots are shown with
their real semantic names (e.g. ${question}, ${insightsJson}).