Catchup: Recent PR Synthesis

Arguments: $ARGUMENTS

Parameters (order-independent, all optional):

• hours — numeric, defaults to 24
• author filter — me (only your PRs), others (everyone except you), or omit for all

PR Data

Merged PRs in the window:

!A="${ARGUMENTS:-}"; H=$(echo "$A" | grep -oE '[0-9]+' | head -1); H=${H:-24}; F=$(echo "$A" | grep -oiE '\b(me|others)\b' | head -1 | tr A-Z a-z); case "$F" in me) AF="author:@me";; others) AF="-author:@me";; *) AF="";; esac; gh pr list --state=merged --search="merged:>=$(date -u -v-${H}H +%Y-%m-%dT%H:%M:%S)Z ${AF}" --json number,title,author,mergedAt,body,additions,deletions,changedFiles --limit 50 2>/dev/null || echo "GH_ERROR"

Open/changed PRs in the window (created or updated):

!A="${ARGUMENTS:-}"; H=$(echo "$A" | grep -oE '[0-9]+' | head -1); H=${H:-24}; F=$(echo "$A" | grep -oiE '\b(me|others)\b' | head -1 | tr A-Z a-z); case "$F" in me) AF="author:@me";; others) AF="-author:@me";; *) AF="";; esac; gh pr list --state=open --search="updated:>=$(date -u -v-${H}H +%Y-%m-%dT%H:%M:%S)Z ${AF}" --json number,title,author,createdAt,updatedAt,body,additions,deletions,changedFiles,labels --limit 50 2>/dev/null || echo "GH_ERROR"

Git stats for the window:

!A="${ARGUMENTS:-}"; H=$(echo "$A" | grep -oE '[0-9]+' | head -1); H=${H:-24}; SINCE=$(date -u -v-${H}H +%Y-%m-%dT%H:%M:%S); COMMITS=$(git log --since="$SINCE" --oneline origin/main 2>/dev/null | wc -l | tr -d ' '); STATS=$(git log --since="$SINCE" --shortstat origin/main 2>/dev/null | grep "insertion\|deletion" | awk '{for(i=1;i<=NF;i++){if($i~/insertion/)ins+=$(i-1);if($i~/deletion/)del+=$(i-1)}} END {printf "%d added, %d removed", ins+0, del+0}'); echo "commits=$COMMITS $STATS"

Project context for understanding what changed:

!cat AGENTS.md

Instructions

You are producing a net-effects briefing — not a PR-by-PR changelog. The reader wants to understand what's different about the codebase now compared to N hours ago.

Check the arguments above to determine the author filter. If others was specified, note in your briefing header that this covers changes by other contributors (not the person running the skill). If me was specified, note it covers only the runner's own changes. If no filter, cover all contributors.

Step 1: Triage

If either PR list shows GH_ERROR, report that gh failed (likely auth or network) and stop.

Scan both PR lists. If there are zero PRs in both lists, say so and stop.

Group PRs into two buckets:

• Landed — merged in the window
• In flight — open PRs created or updated in the window

Step 2: Understand the changes

For each merged PR, read its description carefully. Extract:

• What area of the codebase it touches
• What capability it adds, changes, or removes
• Whether it introduces schema/migration changes, API surface changes, new dependencies, or test infrastructure changes

If a PR description is too terse to understand the scope, fetch the diff summary:

gh pr diff <number> --stat

For particularly important or large PRs, fetch the full diff to understand the actual changes:

gh pr diff <number>

Use your judgment — not every PR needs a full diff read. Focus effort on PRs with large insertion counts, schema changes, or architectural implications.

Step 3: Write the briefing

Produce the following sections. Use concise, direct language. No filler.

---

By the Numbers

A compact stats line at the top of the briefing. Use the git stats and PR data provided above to produce a single line like:

12 commits, 6 PRs merged, +4,713 / -128 lines

Include: commit count, merged PR count, lines added/removed. If an author filter is active, note it (e.g., "by @user" or "by others").

Landed

Group changes by theme/area, not by PR. Merge related PRs into a single narrative. For each theme:

• What changed (capabilities added, removed, or modified)
• Relevant stats if notable (e.g., "new sync phase", "15 new API tests")
• Which PRs contributed and who authored them (parenthetical references, e.g., "#23, #24 — @user")

Themes should be whatever emerges naturally from the changes — examples: "Sync pipeline", "Search & indexing", "CRM / entity resolution", "API endpoints", "MCP tools", "Documentation", "Tooling", "Test coverage". Don't force categories that don't exist.

System Impact

For each theme above, explain in plain language how the system's behavior or capabilities changed at a high level. Write this for someone who hasn't looked at the code — they want to know what the system can do now that it couldn't before, what works differently, or what's more reliable/faster/safer.

Examples of the right level of abstraction:

• "The nightly sync pipeline now halts on critical failures instead of continuing with stale data. If ChromaDB goes down mid-sync, downstream phases won't run with incomplete indexes."
• "Hybrid search now weights BM25 keyword matches more heavily for short queries, improving precision on exact-name lookups without degrading semantic recall for longer questions."
• "The CRM person timeline now includes iMessage and WhatsApp data alongside email and calendar, giving a unified interaction history per person."

Avoid restating the "what changed" section — focus on the so what.

In Flight

For each open PR updated in the window:

• Title and author
• What it's adding/changing
• Current state (draft? review requested? CI passing?)

If there are no open PRs, omit this section.

Review Attention

Optional section — include only if warranted. Flag items like:

• Unusually large PRs that may not have received proportional review depth
• Architectural decisions that are now load-bearing and hard to reverse
• Areas with complex logic that could benefit from a closer look
• Cross-cutting changes that affect many files
• New dependencies introduced

If nothing warrants attention, omit this section entirely.

---

Formatting rules

• Use markdown headers and bullets
• Reference PR numbers as #N (GitHub will auto-link them)
• Don't include full PR descriptions — synthesize
• Keep the total output scannable — aim for a briefing someone can read in 2-3 minutes
• Don't editorialize on code quality or contributor velocity

Escalation

Stop and tell the user when:

• gh commands fail (auth, network, or unexpected errors)
• Either PR list returns exactly 50 results (the limit cap) — suggest a shorter time window
• The date command fails (this skill uses macOS date -v syntax; it won't work on Linux)