launch-summary

Files (1)

• SKILL.md

SKILL.md (first lines)

## Instructions

Generate a **non-developer-friendly launch summary** from merged pull requests across four GitHub repos: `Atllas-Inc/codebase`, `Atllas-Inc/aicc-queues`, `Atllas-Inc/neema-simple-hyzl`, and `Atllas-Inc/pointsgpt`.

The skill takes one argument, the window: `daily` or `weekly`. **If the user gives no argument, use `daily`.** The window sets the timeframe and nothing else. Every later step is shared.

### Step 1: Determine the date range

- `daily`: the last 24 hours, a rolling window ending now.
- `weekly`: the current calendar week, Monday 00:00 UTC through today.

If the user specified a different range, a specific day, or a specific week, use that instead.

### Step 2: Fetch merged PRs from every repo

Only PRs merged **into the repo's default branch** count. The `base:` filter excludes PRs merged into release branches and feature branches.

**Never hardcode `base:master`.** The default branch differs per repo: `codebase` and `aicc-queues` use `master`, `neema-simple-hyzl` and `pointsgpt` use `main`. Read it from `repos/<repo>` so a repo added later needs no edit here.

**Never use `gh pr list`.** It calls the GitHub GraphQL API, which Claude Code sessions block with a 403. Use the REST endpoints below.

`search/issues` filters on merge date server-side, so no `--limit` can drop a PR that was opened long ago and merged inside the window. It returns the author, but not the merger and not the changed files. Step 2 therefore calls `pulls/{n}` for every PR to read `merged_by`. It also calls `pulls/{n}/files` for each repo that ships a phone app, to tell Mobile PRs from App PRs.

Two repos ship one: `codebase` under `apps/atllas-app/`, and `pointsgpt` under `ios/`. The `mobile_prefix` function below holds the prefix per repo. An empty value means no phone app.

`product_name` holds the customer-facing name per repo. The summary is grouped by that name, so every repo needs a row.

`display_name` turns a GitHub login into a person's real name, and caches each lookup in `/tmp/ls_names.tsv` so one person costs one call.

Add a repo by appending it to the `for REPO in ...` list and giving it a `product_name` row. **Never put the repo list in a variable.** This block runs under `zsh`, which does not word-split an unquoted variable, so a list in a variable is read as one repo name.

Run the whole block in one Bash invocation so `SINCE` is computed once:

```bash
set -euo pipefail
WINDOW=daily   # daily | weekly

case "$WINDOW" in
  daily)
    SINCE=$(date -u -v-24H +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || date -u -d "24 hours ago" +%Y-%m-%dT%H:%M:%SZ 2>/dev/null)
    [ -n "$SINCE" ] || { echo "ERROR: could not compute a 24-hours-ago timestamp with this system's date(1). Aborting." >&2; exit 1; }
    HEADER="## 🚀 Daily Launch Summary: $(date -u +%Y-%m-%d)"
    EMPTY="Nothing shipped in the last 24 hours."
    ;;
  weekly)
    WEEK_START=$(date -v-Mon +%Y-%m-%d 2>/dev/null || date -d "last Monday" +%Y-%m-%d 2>/dev/null)
    [ -n "$WEEK_START" ] || { echo "ERROR: could not compute the start of the week with this system's date(1). Aborting." >&2; exit 1; }
    SINCE="${WEEK_START}T00:00:00Z"
    HEADER="## 🚀 Weekly Launch Summary: Week of ${WEEK_START}"
    EMPTY="Nothing shipped this week."
    ;;
esac
UNTIL=""
NAMES=/tmp/ls_names.tsv
: > "$NAMES"
echo "Window starts: $SINCE"
echo "Header: $HEADER"
echo "Empty-result line: $EMPTY"

mobile_prefix() {
  case "$1" in
    codebase) echo '^apps/atllas-app/' ;;
    pointsgpt) echo '^ios/' ;;
    *) echo '' ;;
  esac
}

product_name() {
  case "$1" in
    codebase) echo 'Atllas' ;;
    aicc-queues) echo 'AI Calling' ;;
    neema-simple-hyzl) echo 'Hyzl' ;;
    pointsgpt) echo 'PointsGPT' ;;
    *) echo "$1" ;;
  esac
}

display_name() {
  [ -n "$1" ] || { echo ''; return; }
  local hit
  hit=$(awk -F'\t' -v l="$1" '$1==l {print $2; exit}' "$NAMES")
  if [ -z "$hit" ]; then
    hit=$(gh api "users/$1" --jq '.name // .login')
    printf '%s\t%s\n' "$1" "$hit" >> "$NAMES"
  fi
  echo "$hit"
}

search_prs() {
  gh api -X GET search/issues \
    -f q="repo:$1 is:pr is:merged base:$2 merged:>=$SINCE" \
    -f per_page=100 --paginate \
    --jq '.items[] | {number, title, mergedAt: .pull_request.merged_at, body, author: .user.login, labels: [.labels[].name]}'
}

: > /tmp/ls_prs.jsonl
for REPO in Atllas-Inc/codebase Atllas-Inc/aicc-queues Atllas-Inc/neema-simple-hyzl Atllas-Inc/pointsgpt; do
  NAME="${REPO#*/}"
  BASE=$(gh api "repos/$REPO" --jq .default_branch)
  PRODUCT=$(product_name "$NAME")
  PREFIX=$(mobile_prefix "$NAME")
  echo "$REPO: base $BASE, product $PRODUCT"
  search_prs "$REPO" "$BASE" | while read -r row; do
    n=$(jq -r .number <<<"$row")
    m=false
    if [ -n "$PREFIX" ]; then
      if gh api "repos/$REPO/pulls/$n/files?per_page=100" --paginate --jq '.[].filename' \
           | grep -q "$PREFIX"; then m=true; fi
    fi
    MERGER_LOGIN=$(gh api "repos/$REPO/pulls/$n" --jq '.merged_by.login // ""')
    AUTHOR=$(display_name "$(jq -r .author <<<"$row")")
    MERGER=$(display_name "$MERGER_LOGIN")
    jq -c --arg repo "$NAME" --arg product "$PRODUCT" --argjson m "$m" \
          --arg authorName "$AUTHOR" --arg mergerName "$MERGER" \
      '. + {repo: $repo, product: $product, mobile: $m, authorName: $authorName, mergerName: $mergerName}' <<<"$row"
  done >> /tmp/ls_prs.jsonl
done

jq -s --arg until "$UNTIL" '[.[] | select($until == "" or .mergedAt < $until)]' /tmp/ls_prs.jsonl

From ptrandev/agent-skills, MIT.

More from ptrandev/agent-skills