Standup

Arg syntax is flexible — accept /standup, /standup day, /standup week format=slack, /standup since 2026-04-06 until 2026-04-13, /standup 2w owner=vlt-ai.

Step 2 — Run the collector

Invoke:

bash ~/.claude/skills/standup/collect.sh <RANGE_ARGS> [--owner ORG] [--repo OWNER/NAME] [--user USERNAME]

The collector:

1. Calls gh search commits with the author and date range.
2. For each commit, fetches +additions / -deletions in parallel via gh api repos/OWNER/REPO/commits/SHA.
3. Groups by owner/repo, sums stats, prints structured output.

Default concurrency is 10. For very large ranges (hundreds of commits), pass --no-stats to skip the per-commit API calls, or --parallel 20 to speed up.

Capture the full stdout; that's your data for clustering.

Step 3 — Cluster (skip if format=raw)

Group the per-repo commits into initiatives — coordinated multi-repo work streams.

If a Brain vault is available (check $BRAIN_VAULT, then ~/vlt-workspace/brain/, then ~/brain/):

• Read brain/notes/repo-dependency-graph.md (or any file matching *dependency* / *repo-map* / *service*repo*)
• Use its edges to identify coordinated work: a commit in repo A that bumps a submodule pointing at repo B, or shared commit-message scopes across repos (e.g. feat(scaling) in both thor + hrym)

Without a Brain vault, infer clusters from signals in the data itself:

• Shared conventional-commit scopes across repos (feat(scaling), fix(metrics))
• Submodule-bump commits (chore: update <other-repo> submodule)
• Lines-changed magnitude — big diffs in matched scopes across two repos usually signal a coordinated push
• Dates clumping around the same theme
• Subject-line similarity

Name each cluster with a short initiative title, not a repo name.

Step 4 — Render

Format: standup (default)

## Standup — <date range>

**Shipped**
- <Initiative name>: <one-line what-changed, span the relevant repos>
- ...

**Critical fixes**
- <short list of crashes, regressions, hotfixes>

**Tooling**
- <devex / CLI / infra helpers>

**By the numbers**
<N commits across M repos, +X / -Y lines. Call out the 1-2 dominant multi-repo initiatives.>

**In flight**
- <WIP branches, incomplete POCs, pinned/paused items>

**Watch-outs for the team**
- <renames, removed paths, config changes others will hit>

Keep it tight — a standup is skimmed in under a minute. Lean on the lines-changed data to call out the dominant efforts.

Format: report

Prose paragraphs, one per initiative. Include context ("why we did this"), outcomes, and who/what it affects. Suitable for a written weekly update or manager 1:1 prep. Cite commit counts and line deltas as evidence.

Format: slack

Compact blocks with emoji markers. Short bullets. Include a :chart_with_upwards_trend: line with the commit + line counts. Good for pasting into a team channel.

Format: raw

Print the collector output verbatim. No analysis. Useful for scripting or debugging.

Step 5 — Offer follow-ups

End with a single short line offering relevant follow-ups: narrower/wider range, different format, drill into one initiative, export to file.

Arguments reference

/standup                                    # week, standup format
/standup day                                # just today
/standup yesterday
/standup week                               # explicit
/standup 2w                                 # last 14 days
/standup month                              # last 30 days
/standup since 2026-04-06
/standup since 2026-04-06 until 2026-04-13
/standup week format=slack
/standup day format=report
/standup week owner=vlt-ai                  # restrict to one org
/standup week repo=vlt-ai/thor              # restrict to one repo
/standup week user=someuser                 # someone else's commits
/standup raw                                # skip clustering, show data only

Flags for collect.sh (pass through when the user's ask requires them):

Notes

• If gh is not authenticated, stop and tell the user to run gh auth login — don't try to work around it.
• The collector is deterministic and cheap for a week (~5-10s for ~50 commits with stats). Re-run freely if the user changes the range mid-conversation.
• When clustering, always prefer initiative-centric titles over repo names. "Thor autoscaling foundation" beats "thor + hrym commits".
• If total commits is 0, say so plainly and suggest widening the range or dropping the author filter (--user changes, or different date).
• GitHub's commit search is indexed — commits made seconds ago may not appear. If the user expects a commit that isn't there, mention the indexing lag.
• If you see the 1000-commit cap being hit (rare), suggest narrowing with --owner or a shorter range.
• If a Brain vault is available, cite it briefly (e.g. "clustered per repo-dependency-graph") so the user knows where the grouping came from.