common-groundlisted
Install: claude install-skill Wang-Cankun/cankun-skills
# Common Ground
The user and the agent should hold the same picture of the task: what it is for, where it
stands, and what comes next. The user calls this skill when that picture has split: the agent
talks in terms they cannot follow, or the work has drifted from what they meant. A typed
invocation pauses work that depends on the answer until the user confirms; a probe or sketch
that makes the picture concrete is part of the answer.
## Hand the understanding back
Start from one concrete case in plain words, in the user's own names for things (write it the
way `hai-simplified-technical` would), then widen. Write one message for a reader who was not
watching:
1. **Bottom line**: what the task is for, and what is now true that was not known when the
work started.
2. **What changed**: the premises, plans or judgments the work overturned or confirmed,
measured against what the user asked for.
3. **Evidence, inference, confidence**: verified facts with where to check them, your
inferences, and your confidence, kept visibly apart.
4. **Decisions**: the choices that are the user's, each with a concrete example, options and a
recommendation, or "none".
5. **Defaults taken**: choices made on the user's behalf that they may override.
Cut what would not change the user's next decision and link to detail instead. The message
works when the user can decide without a follow-up question.
## Build a shared mental model
Investigate the relevant context and explain the situat