diff --git a/skills/pr-digest/SKILL.md b/skills/pr-digest/SKILL.md
new file mode 100644
index 0000000..4555a5f
--- /dev/null
+++ b/skills/pr-digest/SKILL.md
@@ -0,0 +1,112 @@
+---
+name: pr-digest
+description: Landscape view of open pull requests across the Mu2e offline repos — who authored what, review state, CI state at the current head, and which PRs need your attention. Use when asked for a PR summary, PR status, "what's open on Offline", or what needs reviewing. Read-only.
+compatibility: Requires an authenticated `gh` CLI and network access to github.com. Makes no writes.
+metadata:
+  version: "1.0.0"
+  last-updated: "2026-08-08"
+---
+
+# PR Digest
+
+## Purpose
+
+Answer "what is the state of the PR queue?" — as opposed to
+`reviewing-pull-requests`, which answers "is this one PR any good?".
+
+Use it for: a standup-style summary, deciding what to review next,
+finding PRs whose head moved after you reviewed them, and spotting CI
+that is red or was never run at the current head.
+
+## Read-only
+
+Every call this skill makes is a GET. It never posts a review, never
+comments, never triggers a build, never edits or merges anything. That
+is the point: a digest is cheap to trust precisely because it cannot
+change anything. If a PR in the digest needs action, hand it to
+`reviewing-pull-requests` — do not act from here.
+
+## Scope
+
+The nine Mu2e offline repos: `Offline`, `Production`, `EventNtuple`,
+`EventDisplay`, `DQM`, `Tutorial`, `PassN`, `RefAna`, `ArtAnalysis`.
+Passing any other repo is a hard error, not a silent skip.
+
+Open, non-draft PRs only. Drafts are excluded — they are not asking for
+review yet.
+
+## Usage
+
+```
+python3 scripts/pr_digest.py               # all nine repos
+python3 scripts/pr_digest.py Offline       # one repo
+python3 scripts/pr_digest.py Offline Production
+python3 scripts/pr_digest.py --json        # machine-readable
+```
+
+Roughly 15 s for all nine repos (~20 API calls for ~12 open PRs).
+Exit status is 1 if any repo failed to report, 0 on a clean run — so a
+caller can distinguish a partial picture from a complete one.
+
+## What it reports
+
+**NEEDS ATTENTION** — only PRs with at least one flag, most-flagged
+first. Nothing here means nothing is flagged, which is stated explicitly
+rather than left as an empty section.
+
+| flag | meaning | why it matters |
+|---|---|---|
+| `your review is STALE` | you reviewed at sha X, head is now Y | your findings describe code that no longer exists; the author pushed after you |
+| `CI RED at head` | a check reports FAILURE at the current head | names the failing contexts, so you can tell a real break from an infra blip |
+| `no CI at head` | Offline/Production PR with no check at this head | the green you see may belong to an older commit — see `reviewing-pull-requests` §Triggering a CI Build |
+| `never reviewed by anyone` | no `reviewDecision` and no review by you | the queue's actual backlog |
+| `merge CONFLICT` | `mergeable == CONFLICTING` | blocked regardless of review state |
+
+**ALL OPEN** — one row per PR: repo#number, author, age in days, CI
+mark, overall review decision, and your own review state.
+
+The `you` column is the one to read: `@head` means your review covers
+the current head, a sha8 means your latest review is at that older
+commit, `-` means you have not reviewed it.
+
+CI marks: `ok` green, `FAIL` red, `run` pending, `--` no checks.
+
+**INCOMPLETE** — any repo that did not report cleanly is listed by name
+with its error, and the totals are labelled a lower bound. A repo that
+errors is never silently rendered as zero PRs.
+
+## Interpreting `--` in the CI column
+
+`--` is normal and not a finding on `EventNtuple`, `EventDisplay`,
+`DQM`, `Tutorial`, `PassN`, `RefAna` and `ArtAnalysis` — FNALbuild does
+not watch those repos, so there is no CI to be missing. The digest only
+raises "no CI at head" for `Offline` and `Production`, where the absence
+is genuinely informative.
+
+## Implementation notes
+
+Two `gh` quirks drove the call structure. Do not "optimize" them away:
+
+- **`gh pr list --json latestReviews` returns `commit.oid` as an empty
+  string.** The whole point of the `you` column is comparing your
+  review's commit against the live head, so the digest makes one
+  `gh api .../pulls/<N>/reviews` call per PR to get `commit_id`. There is
+  no way to get it from the list endpoint.
+- **`latestReviews` also carries the full review `body`.** On a PR with
+  a long review that is ~15 KB of prose per PR, for a field the digest
+  never displays. `latestReviews` is deliberately absent from the
+  requested field list.
+
+A PR is treated as covered if *any* of your reviews sits at the current
+head, not merely the most recent one — posting a 🟡 comment after an
+earlier 🔴 at the same sha should not read as stale.
+
+## What it deliberately does not do
+
+- **No quality judgement.** It reports that CI is red, not why. It
+  reports that a PR is unreviewed, not whether it is any good.
+- **No unanswered-question detection.** Deciding whether a reviewer's
+  inline question was actually answered needs to read the thread and
+  judge it — that is review work, and it belongs in
+  `reviewing-pull-requests`, not in a digest that must stay cheap.
+- **No writes.** See above.
diff --git a/skills/pr-digest/scripts/pr_digest.py b/skills/pr-digest/scripts/pr_digest.py
new file mode 100755
index 0000000..64b9cfb
--- /dev/null
+++ b/skills/pr-digest/scripts/pr_digest.py
@@ -0,0 +1,201 @@
+#!/usr/bin/env python3
+"""Read-only landscape view of open PRs across Mu2e offline repos.
+
+Makes no writes of any kind. Every gh call is a GET.
+
+Usage:
+    pr_digest.py [repo ...]      # default: all nine skill-scope repos
+    pr_digest.py --json          # machine-readable, no table
+    pr_digest.py Offline         # one repo
+"""
+
+import json
+import subprocess
+import sys
+from datetime import datetime, timezone
+
+REPOS = ["Offline", "Production", "EventNtuple", "EventDisplay", "DQM",
+         "Tutorial", "PassN", "RefAna", "ArtAnalysis"]
+
+# FNALbuild only watches these two; elsewhere "no CI" is normal, not a finding.
+FNALBUILD_REPOS = {"Offline", "Production"}
+
+PR_LIMIT = 100  # if a repo ever exceeds this we say so rather than truncate silently
+
+LIST_FIELDS = ("number,title,author,createdAt,updatedAt,headRefOid,"
+               "reviewDecision,mergeable,mergeStateStatus,statusCheckRollup,url,isDraft")
+
+
+def gh(args):
+    """Run a gh command, returning (stdout, error_or_None). Never raises."""
+    try:
+        p = subprocess.run(["gh"] + args, capture_output=True, text=True, timeout=120)
+    except (OSError, subprocess.TimeoutExpired) as e:
+        return "", f"{type(e).__name__}: {e}"
+    if p.returncode != 0:
+        return "", (p.stderr or "").strip().splitlines()[-1] if p.stderr else f"exit {p.returncode}"
+    return p.stdout, None
+
+
+def me():
+    out, err = gh(["api", "user", "--jq", ".login"])
+    if err:
+        sys.exit(f"FATAL: cannot determine gh user ({err}). Is gh authenticated?")
+    return out.strip()
+
+
+def ci_summary(rollup, repo):
+    """Collapse statusCheckRollup into (state, detail). State in
+    green/red/pending/none."""
+    if not rollup:
+        return ("none", "no checks at head" if repo in FNALBUILD_REPOS else "n/a")
+    fail, pend, ok = [], [], 0
+    for c in rollup:
+        # StatusContext uses state; CheckRun uses conclusion/status.
+        st = (c.get("state") or c.get("conclusion") or c.get("status") or "").upper()
+        name = c.get("context") or c.get("name") or "?"
+        if st in ("FAILURE", "ERROR", "TIMED_OUT", "CANCELLED"):
+            fail.append(name)
+        elif st in ("PENDING", "IN_PROGRESS", "QUEUED", "EXPECTED", ""):
+            pend.append(name)
+        else:
+            ok += 1
+    if fail:
+        return ("red", ", ".join(sorted(fail)[:3]) + ("…" if len(fail) > 3 else ""))
+    if pend:
+        return ("pending", ", ".join(sorted(pend)[:3]) + ("…" if len(pend) > 3 else ""))
+    return ("green", f"{ok} checks")
+
+
+def my_review(repo, num, user, head8):
+    """Latest review by `user`: (state, sha8, at_head) or None. One GET."""
+    out, err = gh(["api", f"repos/Mu2e/{repo}/pulls/{num}/reviews", "--paginate",
+                   "--jq", f'.[] | select(.user.login=="{user}") '
+                           '| "\\(.state)\\t\\(.commit_id[0:8])\\t\\(.submitted_at)"'])
+    if err:
+        return ("ERROR", err[:40], False)
+    rows = [r.split("\t") for r in out.strip().splitlines() if r.strip()]
+    if not rows:
+        return None
+    rows.sort(key=lambda r: r[2])          # chronological; last is most recent
+    state, sha, _ = rows[-1]
+    # Any review at the live head counts as covered, even if an older one was later.
+    at_head = any(r[1] == head8 for r in rows)
+    return (state, sha, at_head)
+
+
+def collect(repos, user):
+    prs, errors = [], []
+    for repo in repos:
+        out, err = gh(["pr", "list", "--repo", f"Mu2e/{repo}", "--state", "open",
+                       "--limit", str(PR_LIMIT), "--json", LIST_FIELDS])
+        if err:
+            errors.append(f"{repo}: {err}")
+            continue
+        try:
+            items = json.loads(out or "[]")
+        except json.JSONDecodeError as e:
+            errors.append(f"{repo}: unparseable gh output ({e})")
+            continue
+        if len(items) >= PR_LIMIT:
+            errors.append(f"{repo}: hit the {PR_LIMIT}-PR cap; list may be truncated")
+        for it in items:
+            if it.get("isDraft"):
+                continue
+            head8 = (it.get("headRefOid") or "")[:8]
+            ci, ci_detail = ci_summary(it.get("statusCheckRollup"), repo)
+            mine = my_review(repo, it["number"], user, head8)
+            created = datetime.fromisoformat(it["createdAt"].replace("Z", "+00:00"))
+            prs.append({
+                "repo": repo, "number": it["number"], "title": it["title"],
+                "author": (it.get("author") or {}).get("login", "?"),
+                "age_days": (datetime.now(timezone.utc) - created).days,
+                "head": head8, "url": it["url"],
+                "review_decision": it.get("reviewDecision") or "",
+                "mergeable": it.get("mergeable") or "",
+                "merge_state": it.get("mergeStateStatus") or "",
+                "ci": ci, "ci_detail": ci_detail,
+                "my_review_state": mine[0] if mine else "",
+                "my_review_sha": mine[1] if mine else "",
+                "my_review_at_head": bool(mine and mine[2]),
+            })
+    return prs, errors
+
+
+def attention(pr, user):
+    """Reasons this PR wants a human. Order matters: most actionable first."""
+    out = []
+    if pr["my_review_state"] and not pr["my_review_at_head"]:
+        out.append(f"your review is STALE (reviewed {pr['my_review_sha']}, head {pr['head']})")
+    if pr["ci"] == "red":
+        out.append(f"CI RED at head — {pr['ci_detail']}")
+    if pr["ci"] == "none" and pr["repo"] in FNALBUILD_REPOS:
+        out.append(f"no CI at head {pr['head']}")
+    if not pr["review_decision"] and not pr["my_review_state"]:
+        out.append("never reviewed by anyone")
+    if pr["mergeable"] == "CONFLICTING":
+        out.append("merge CONFLICT")
+    return out
+
+
+CI_MARK = {"green": "ok", "red": "FAIL", "pending": "run", "none": "--"}
+
+
+def render(prs, errors, user, repos):
+    W = 78
+    print(f"Mu2e PR digest — {datetime.now().strftime('%Y-%m-%d %H:%M')} — as {user}")
+    print(f"{len(repos)} repo(s), {len(prs)} open non-draft PR(s)")
+    print("=" * W)
+
+    flagged = [(p, attention(p, user)) for p in prs]
+    flagged = [(p, r) for p, r in flagged if r]
+    if flagged:
+        print("\nNEEDS ATTENTION")
+        for p, reasons in sorted(flagged, key=lambda x: (-len(x[1]), x[0]["repo"])):
+            print(f"  {p['repo']}#{p['number']}  {p['title'][:52]}")
+            for r in reasons:
+                print(f"      - {r}")
+            print(f"      {p['url']}")
+    else:
+        print("\nNEEDS ATTENTION: nothing flagged.")
+
+    print("\nALL OPEN")
+    hdr = f"  {'PR':<20} {'author':<16} {'age':>4} {'CI':<5} {'review':<16} {'you':<10} title"
+    print(hdr)
+    print("  " + "-" * (len(hdr) - 2))
+    for p in sorted(prs, key=lambda x: (x["repo"], -x["number"])):
+        you = ("@head" if p["my_review_at_head"]
+               else (p["my_review_sha"] if p["my_review_state"] else "-"))
+        dec = (p["review_decision"] or "-").replace("CHANGES_REQUESTED", "CHANGES_REQ")
+        print(f"  {p['repo']+'#'+str(p['number']):<20} {p['author'][:16]:<16} "
+              f"{str(p['age_days'])+'d':>4} {CI_MARK[p['ci']]:<5} {dec[:16]:<16} "
+              f"{you:<10} {p['title'][:28]}")
+
+    if errors:
+        print("\nINCOMPLETE — these repos did not report cleanly:")
+        for e in errors:
+            print(f"  ! {e}")
+        print("  Treat counts above as a lower bound, not a complete picture.")
+
+
+def main():
+    argv = [a for a in sys.argv[1:] if a != "--json"]
+    as_json = "--json" in sys.argv[1:]
+    repos = argv or REPOS
+    bad = [r for r in repos if r not in REPOS]
+    if bad:
+        sys.exit(f"FATAL: not a skill-scope repo: {', '.join(bad)}\n"
+                 f"Known: {', '.join(REPOS)}")
+    user = me()
+    prs, errors = collect(repos, user)
+    if as_json:
+        print(json.dumps({"user": user, "repos": repos, "prs": prs,
+                          "errors": errors}, indent=2))
+    else:
+        render(prs, errors, user, repos)
+    # Exit 1 if any repo failed, so a caller can tell a partial run from a clean one.
+    return 1 if errors else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())