standup

---
name: standup
description: Use when the user asks "what did I do today", "what have you done today", "what did corgi do while I was out", "what did I do yesterday", "write my standup", "what landed this week". Runs `corgi agent today` / `corgi agent standup`, reads it back, offers to post. NOT for "what's blocked" (tracker).
---

# Standup

**Done when:** the standup text is printed (and posted, when asked), built from what the day's git, tracker and sessions actually show.

Read `../_shared/conventions.md` first.

`corgi agent today` already knows what every account's Claude Code was asked, what
landed in git under each registered workspace, and - the part nothing else records -
what the unattended watch did on its own. This skill runs it, reads it, and turns it
into words a person would say.

## Run

- "Today", "what have you done today", "what did corgi do while I was out" →
  `corgi agent today`. Its window is midnight, not a rolling 24h, so the answer does
  not drift into yesterday evening. It prints a headline first: commits, pull requests
  the watch opened, runs still going, what is waiting.
- A window that is not today → `corgi agent standup --since 48h` (a Monday's
  "yesterday"), `--since 72h` (the weekend), `--since 168h` (this week). Same data,
  rolling window. `corgi agent today --since 8h` works too.
- The user asked for prose ("write it", "summarize", "three sentences") → add
  `--write`. That hands the raw list to `claude -p` and prints a few sentences; without
  it nothing leaves the machine, so say which mode you used.
- Empty output → tracking may be off: point at `corgi agent track enable`, then fall
  back to `git log --since=<window> --oneline` per repo of the current workspace.
- In a chat with the MCP server, `corgi_today` returns the same thing without a shell.

## Read it back

Group by workspace as the command does. Commits are facts (keep the subject, drop the
hash); prompts are headlines (what was asked, not the wording). Turn both into:

```
<workspace>
  done     <what landed, one line per PR or commit cluster>
  corgi    <what the unattended watch did, and the PR it opened - only when it did>
  doing    <what the last prompts were about, not yet committed>
  next     <what the open branches imply>
  blocked  <only when sessions or brief show a wait or a limit>
```

Say plainly which work was yours and which the watch did on its own - a pull request
you never opened is the one line a standup must not blur. A `waiting for a free slot`
list is `blocked`: a cap or quiet hours held those back, and a manual
`corgi agent watch run` picks them up.

Three to six lines per workspace; only the cwd workspace for a team standup, all of
them for "everything".

Cross-check the tracker only when ticket keys appear in the commits: pull those keys'
status through the `tracker` skill and mark drift ("ticket still In Progress, PR
merged").

## Offer to post

Ask once: "post this?" with the destinations that exist: the tracker (a comment on
the cycle or the standup issue, through the `tracker` skill's write gate), the chat
the daemon already notifies (`corgi agent digest --send` sends today's digest there),
or a paste-ready block for Slack. Never post without the answer, never add
attribution, never make the facts sound bigger than the diff.

## Do not

- Redo the git archaeology yourself when the command returned data.
- Read prompt history beyond what the command prints.

## See also

- **[`summary`](../summary/SKILL.md)** - when the question is "where does my batch stand
  **now**" rather than "what did I do". That one re-reads every PR/MR and ticket live and
  groups by ticket; this one narrates a period of work for people who were not here.