cmd-code-what

What

Give the user a high-signal, low-noise snapshot of what's happened in this session so they can re-enter the loop without scrolling back.

Output format

Reply with 3-5 bullet points only. No preamble, no closing summary, no headers, no follow-up questions. Each bullet must be:

- **1-3 word label**: explanation in less than 120 characters

Hard rules:

• 3-5 bullets, no more, no less in spirit (4 is the sweet spot)
• Label: 1-3 words, bolded, title-case or kebab-case, names a concrete thing (file, action, decision, blocker)
• Explanation: a single clause, < 120 characters including the label, present tense or past tense — never future tense ("will", "going to")
• No nested bullets, no code fences, no links, no emoji unless one is genuinely load-bearing
• Skip anything the user already knows from the last 1-2 turns — focus on what they'd miss if they just walked back to the terminal

What to include (in priority order)

1. Current focus — what file/task is actively being worked on right now
2. Most recent concrete change — what just got written, edited, run, or decided
3. Blocker or open question — anything waiting on the user, or that's stuck
4. Next step — the very next action the agent intends to take, if obvious
5. Notable side effects — files created, branches touched, processes started, only if relevant

What to leave out

• Restating the original user request verbatim
• Tool-call narration ("I ran ls, then grep, then…")
• Anything that happened more than ~10 turns ago unless it's still load-bearing
• Hedging, qualifiers, or meta-commentary about the summary itself

Example

User: /cmd-what

Good response:

- **Editing**: refactoring `auth/session.py` to drop the legacy cookie path
- **Tests**: 14/16 passing, 2 failing on token refresh edge case
- **Blocked**: need to know if expired refresh tokens should silently re-auth or 401
- **Next**: wire the chosen behavior into `refresh_token()` once you decide

Bad response (too verbose, narrates tool calls, future-tense, missing labels):

- I started by reading the auth module and then I grepped for cookie usage
- I'm planning to refactor the session handling but first I want to check with you
- The tests will probably need updates too