fix: restore CVEs report column heading

Only this event entry committed; pre-existing uncommitted orchestrator WIP left in the worktree untouched.

.claude/skills/cve-check/SKILL.md

@@ -95,8 +95,9 @@ For each recipe produce **one window per image**: `current pinned tag → newest
 the sidecars (redis, postgres, nginx …), not just the app — a sidecar bump is where discourse's only
 CRITICAL came from, and an image with no window is not counted at all.
 
-- **No upgrade available** → the recipe is `UPTODATE`; its CVE count is **`0`**, not `?`. There is
-  nothing an upgrade could fix. Record it and move on.
+- **No upgrade available** → the recipe is `UPTODATE`; its **CVEs fixed by an available upgrade**
+  count is `0`. This does not assert that the running image has no known CVEs. If advisory sources
+  failed or are blind, use `?` and say why.
 - **No output at all is NOT "no upgrade".** An abra call that times out, FATAs, or prints nothing
   leaves the recipe **unverified** — treat it as a distinct outcome, never fold it into up-to-date.
   Re-run it, and if it still yields nothing, resolve the versions by direct registry check (box item 4).
@@ -111,9 +112,17 @@ python3 cc-ci-plan/audit-sources.py --security-sources
 A recipe whose sources yield **no CVE data at all** cannot produce a meaningful `0` — nothing was
 measured, the same way a missing registry file cannot. Render those as **`?`**, not `0`.
 
-**The fleet is currently at zero such recipes.** The last two — `mattermost-lts` (empty advisory
-feed, client-side-rendered bulletins) and `mumble` (nothing published anywhere) — were fixed by
-declaring an NVD CPE in their registry:
+Do not assume that the fleet has zero blind recipes. Run the audit every sweep and inspect source
+statuses; an HTTP error or rate limit is not evidence of no CVEs. `rustdesk-server` is currently
+blind because its upstream GitHub repo publishes no security advisories and no other usable source
+has been verified; mark it `?` until that changes.
+
+Check `test -s /srv/cc-ci/.github-token` before a fleet scan. Without that read-only token,
+GitHub's anonymous API limit can be exhausted partway through the fleet; HTTP 403 results are
+failed sources, so the affected CVE counts remain unknown.
+
+Two older blind spots — `mattermost-lts` and `mumble` — were addressed by declaring NVD CPEs in
+their registries:
 ```
 - nvd-cpe: mattermost-team-edition = cpe:2.3:a:mattermost:mattermost_server:*:*:*:*:*:*:*:*
 ```

.claude/skills/recipe-report/SKILL.md

@@ -30,7 +30,9 @@ keeps every weekly edition looking the same regardless of which model writes the
 2. **Survey.** `python3 /srv/cc-ci/cc-ci-plan/recipe-report.py survey <DATE> > /tmp/survey.json`. This
    gives you: the raw `/upgrade-all` summary; for every recipe mirror its open PRs + the `cc-ci/testme`
    verdict; and each recipe's **per-recipe upgrade notes** (`upgrade_notes_md` — the breaking-change /
-   **CVE / security** analysis the upgrade did). Read it all.
+   **CVE / security** analysis the upgrade did). Read it all. Also run
+   `python3 /srv/cc-ci/cc-ci-plan/audit-sources.py --security-sources`; a recipe with no verified
+   security source or a failed source must be called out, not presented as a clean CVE scan.
 
 3. **Review & classify.** The page order is: **short lead → the full wire table (priority-sorted) →
    Addendum → Security Bulletin → per-recipe "What changed".** Same newspaper style; this order.
@@ -46,9 +48,13 @@ keeps every weekly edition looking the same regardless of which model writes the
        lists **failed sources**, the count is **not** authoritative: render the cve cell as `?`
        (unknown), never `none` — a blank that reads as "clean" is exactly how two CVSS-9.8 gitea
        RCEs were reported as "none" on 2026-08-07.
-     - **`?` must stay RARE — it means "we tried and could not tell", not "we didn't look".** Use it
-       ONLY when a scan ran and reported genuinely failed sources, **or when the scan block says
-       COUNT UNKNOWN**. In that case the scan's `0` means *not determined*: publish `?` and say so in
+     - The table's **CVEs** column counts known CVEs addressed by the available recipe upgrade.
+       A zero for an up-to-date recipe means there is no upgrade window to credit; it does not
+       certify that the running images are free of vulnerabilities. If an upgrade exists but its
+       advisory scan is absent, a source failed, or an image window was unmatched, use `?`.
+     - **`?` means the count is not verified.** Use it when an upgrade has no scan, a source failed,
+       an image window was unmatched, or the scan block says COUNT UNKNOWN. In that case publish
+       `?` and say why in
        the notes; publishing `0` would assert a clean bill of health nothing supports. Note a
        **version-scheme change is no longer a reason for `?`** — the scan resolves semver→calver jumps
        (discourse 3.5.3 → 2026.7.1) by falling back to advisory publish dates and reports a real number.

.claude/skills/upgrade-all/SKILL.md

@@ -143,12 +143,31 @@ ssh cc-ci 'export PATH=/run/current-system/sw/bin:$PATH; R=<r>; \
   script -qec "abra recipe upgrade $R -m -n" /dev/null; \
   git -C ~/.abra/recipes/$R stash pop >/dev/null 2>&1 || true'
 ```
-Build `RECIPES_TO_UPGRADE` = recipes with **≥1 available upgrade** (after the stash, so the untracked
-overlay doesn't count as dirty). Others go to `SKIPPED_UPFRONT` with a reason (`dirty-worktree` only for
-real TRACKED edits, `up-to-date`, `not-fetchable`). **A `tag and digest … not supported` parse FATA is
-NOT `not-fetchable`** — per box item 4, catch it and resolve availability via the direct upstream-tag
-check; only `not-fetchable` when the mirror genuinely can't be fetched (network/auth), never for a
-digest-pinned image.
+**Mandatory registry cross-check for EVERY recipe and EVERY image, even if abra says current:** after
+reconciling the mirror and refreshing the recipe checkout, run
+`python3 /srv/cc-ci/cc-ci-plan/resolve-images.py <recipe> --recipe-dir /root/.abra/recipes`
+on cc-ci. Read the JSON `images`, `upgrades_available`, `unresolved`, and `all_resolved` fields.
+The candidate list is the **union** of abra's upgrades and the resolver's upgrades. CryptPad's
+`version-YYYY.M.D` app tag is a known counterexample: abra only reports a parse warning, while the
+resolver finds the newer image. A warning is never evidence that an image is current. Inspect both
+`newest_same_shape` and `newest_within_major`; a newer major is still a candidate for planning,
+though it must not be applied without a compatibility check.
+
+Build `RECIPES_TO_UPGRADE` from every recipe with at least one candidate. Only label a recipe
+`up-to-date` if abra completed successfully **and** the resolver checked every relevant image
+(`all_resolved: true`) **and** neither found a candidate. If an image is unresolved, templated,
+floating, or lacks comparable tags, investigate it directly and record the evidence. Otherwise
+record `unverified` with the image and reason; never fold it into `up-to-date`. `dirty-worktree`
+applies only to real tracked edits. A tag+digest parse failure is not `not-fetchable`.
+
+Run `python3 /srv/cc-ci/cc-ci-plan/audit-sources.py --security-sources` once for the full inventory
+and include any `NO VERIFIED CVE SOURCE` recipes or failed advisory sources in the weekly summary.
+Do not describe their CVE count as clean. This audit is a coverage check, not a substitute for
+`advisory-scan.py` on each actual upgrade window.
+Before the audit, check `test -s /srv/cc-ci/.github-token`. If it is missing, record that GitHub
+advisory checks are subject to the anonymous 60-request/hour limit; any resulting HTTP 403 is a
+failed source and must become `?` in the report. Never replace it with `0` or silently retry from
+model memory. The file is a read-only GitHub token location; never print its contents.
 
 **Reconcile every candidate's mirror during the survey** (even the up-to-date ones), so merged-upstream
 PRs are closed and every mirror `main` tracks true upstream — fleet-wide, not just where there's a new

cc-ci-plan/JOURNAL.md

@@ -1331,3 +1331,82 @@ Weekly run 2026-10-02 complete: 20 considered · 9 upgraded (PRs opened/extended
 Key highlights: gitea upstream shipped its first 2-digit stable (28.0.0-rootless, 2026-09-29 — BREAKING git-egress/actions-retention notes documented in PR #10); ghost 6.67.0 fixes HIGH CVE-2026-84383; nulls held at mysql 8.4 / mariadb 12.3 / postgres 13/15 lines per precedent; matrix-synapse extended to synapse 1.162.0 + MAS 1.26.0; immich bumped v3.2.4 (rc.3.3 declined). Operational: recipe-maintainers/gitea mirror had LOST its master branch (abra whole-recipe FATA) — repaired both refs to upstream HEAD f167b73; reconcile-upstream.sh should learn to recreate a missing upstream-default branch ref, not only sync main. 2 targets (gitea, n8n 2.42.2) were caught in follow-up passes after the initial worklist mis-derivation — nothing left unprocessed. Sweep start: 4 leaked volumes + 46.3 GB images reclaimed (disk 70%->40%); end-of-run reap 0 leaked dev deploys; disk 65% (51G free).
 Per-recipe logs: .cc-ci-logs/upgrades/<recipe>-upgrade-2026-10-02.md (10 files incl. survey).
 NEVER MERGED anything; all PRs await operator review + the operator's `abra recipe release <recipe> -x/y/z` after each merge.
+
+### Event 2026-10-05 ~17:45 UTC — gitea PR #10 stale test fixed (cc-ci PR #44, verified GREEN, NOT merged)
+Operator invoked --with-tests for gitea PR #10 (28.0.0-rootless, RED on test_lfs_roundtrip run #31).
+Diagnosis: stale test — `_api()` caught only HTTPError so one urlopen(timeout=20) socket timeout
+during the 28.0.0 post-restart migration window escaped the poll loop; poll also hard-coded 120s
+deadline vs recipe_meta HTTP_TIMEOUT=600 (STYLE.md §5). Fix: never-raising _api (mirrors
+lifecycle.http_fetch) + deadline from recipe_meta.HTTP_TIMEOUT; assertions untouched.
+cc-ci PR #44 (branch fix/gitea-lfs-restart-wait) also carries the two PRE-EXISTING main lint
+failures (warm.py ruff format; acme-dns.nix statix W201, eval-identical restructure) that had made
+push build #33 RED and would have blocked merge — build #34 now success, lint.sh PASS.
+Verified: full cold suite on host worktree (e83cd46) vs recipe head 91e42c8 → 5/5 tiers pass,
+test_lfs_roundtrip PASSED; unit sample 315 pass / 4 fail = identical pre-existing failures on main.
+Pair note posted on both PRs: merge #44, then re-!testme #10. Nothing merged.
+
+### Event 2026-10-05 ~18:00 UTC — wrong-recipe card screenshot root-caused; runner fix #45 merged; gitea !testme re-run GREEN
+Operator's !testme on gitea PR #10 (drone build 35) showed a custom-html-tiny screenshot in the card.
+Root cause: drone build numbers restarted when its DB was re-created (2026-09-27) while
+/var/lib/cc-ci-runs still holds old-era dirs (numbered to 1377) — build 35 collided with a May-31
+run 35 whose fossil screenshot.png (12950 B = exactly custom-html-tiny's real shot) was embedded
+because screenshot.capture()'s hook branch SKIPPED the snap when out_path existed (no hook saves
+itself). Run itself was genuinely gitea level 5. Fix (cosmetics-only, R7): results.fresh_artifact_dir()
+wipes owned artifacts at run start + capture() always snaps; 4 unit tests (non-vacuous: capture test
+fails on main). Adversary verification PASS (independent fresh clones + own E2E collision test).
+cc-ci PR #45 merged (squash, 35ba9976); push self-tests 38/39 green; lint gate now green on main
+(#44 had already fixed the two pre-existing lint failures).
+Merged #44 first (operator informed; verified earlier via full cold harness run).
+Re-ran !testme on gitea PR #10 → drone build 40: GREEN, level 5, card links /runs/40/, screenshot
+is a fresh PNG of the run's own gitea (footer 1.24.2 = the upgrade-tier BASE deploy, the last-green
+warm canonical — pre-existing capture timing, correct behavior, only freshness was broken before).
+Gitea PR #10 now fully green incl. the fixed LFS test (uses merged #44 test). NOTHING else merged;
+recipe PR #10 + unrelated PR #4 still await operator merge + `abra recipe release gitea -x`.
+Cleanup: host /tmp/v /tmp/r, local worktrees, test-file mis-commit in orch cc-ci clone reverted.
+### Event 2026-10-23 04:15 UTC — weekly /upgrade-all run complete (DEFAULT mode, nothing merged)
+Run 2026-10-23: 19 considered · 9 upgraded with !testme GREEN · 2 PRs open pending the warm-canonical
+minio-prepull fix (lasuite-docs PR #11, lasuite-drive PR #8 — !testme RED at the upgrade BASE prepull:
+warm canonicals 0.4.0+v5.4.1 / 0.12.0+v0.21.0 pin the REMOVED docker.io/minio/minio; PR comments
+explain; operator: advance those canonicals then re-post !testme) · 0 failed · 8 skipped up-to-date
+(cryptpad discourse drone hedgedoc keycloak mailu mumble wordpress — several had stale survey data;
+each re-verified against true upstream before skipping). GREEN PRs: bluesky-pds #6 · custom-html #10 ·
+custom-html-tiny #9 · ghost #8 · gitea #11 · immich #6 · lasuite-meet #12 · matrix-synapse #8 · n8n #9.
+Summary: .cc-ci-logs/upgrades/upgrade-all-2026-10-23.md (per-recipe logs alongside). Step-0 sweep
+reclaimed 19GB images (disk 61%->49%); end-of-run reap 0 leaked dev deploys.
+NEVER MERGED anything; PRs await operator review + `abra recipe release` after each merge.
+### Event 2026-10-30 02:00 UTC — weekly /upgrade-all run complete (DEFAULT mode, nothing merged)
+Run 2026-10-30: 19 considered · 12 upgraded with !testme GREEN · 1 PR open pending its stale warm-canonical
+base (lasuite-docs PR #11 — !testme RED at the base INSTALL rung: canonical 0.4.0+v5.4.1 references the
+removed collab_ss secret; last week's minio blocker is already fixed via #47; head linted + live 12/12;
+operator: seed a v6-cold canonical, re-post !testme) · 1 failed (mailu PR #9 — restore tier RED 3x on a
+PRE-EXISTING upstream defect: main's new backup support lacks a restart post-hook so /data/main.db is
+never restored; PR + operator comment with options) · 5 skipped (gitea + lasuite-meet were already merged
+upstream by their turn; custom-html-tiny drone mumble up-to-date). GREEN PRs: bluesky-pds #6 · cryptpad #6
+(2026.5.1→2026.10.0 breaking server rewrite handled) · custom-html #10→11 (stale nginx digest re-pin) ·
+discourse #12 · ghost #8 · hedgedoc #3 (pgautoupgrade 16→18 in-place) · immich #7 (pgvectors 0.3.0
+same-scheme re-tag) · keycloak #13 (mariadb 12.3→13.0 + AUTO_UPGRADE) · lasuite-drive #8 (+collabora +
+onlyoffice direct checks) · matrix-synapse #8 · n8n #10 · wordpress #3.
+Summary: .cc-ci-logs/upgrades/upgrade-all-2026-10-30.md (per-recipe logs alongside). Step-0 sweep
+reclaimed 66.76GB images + orphan stacks (disk 92%->49%); end-of-run reap 0 leaked dev deploys.
+NEVER MERGED anything; PRs await operator review + `abra recipe release` after each merge.
+### Event 2026-10-30 19:00–21:30 UTC — rustdesk-server one-off upgrade + FULL ENROLLMENT (orchestrator)
+One-off upgrade requested from scratch: public mirror `recipe-maintainers/rustdesk-server` created +
+synced; recipe PR https://git.autonomic.zone/recipe-maintainers/rustdesk-server/pulls/1 (1.1.12→1.1.16,
+branch upgrade-2e65135, unmerged — operator merges). Upstream checked (low-churn, stray 1.4.5 docker
+tags = aborted push; CVE fix POV: UDP punch-hole amplification GH #670 + mio bump; scan 0 CVE-ids,
+client-side CVEs fixed in client 1.4.9). Dev-deploy converged clean twice, then !== enrollment: tests
+authored (status-key health sidecar overlay compose.ccci.yml serving the private-tool-unreachable
+real hbbs public key; ports 21115-21119 + UDP; ws 101 on 21118/19; identity-key cross-source assert;
+punch-hole-amplification regression test w/ version-gate), committed d71a53c, cc-ci PR #48.
+Verify: run 1 RED (custom tier) — hand-rolled version read returned None so the punch-hole gate
+didn't skip on the 1.1.12 head; fixed via harness lifecycle.deployed_identity + loud fail on
+unparsable tag (1653582, f5f11cd); runs 2+3 GREEN level 5/5 (recipe main with gate skipping, AND
+the upgrade head with the punch-hole test passing actively). PR #48 evidence-commented + MERGED
+(200); --cc-ci-only deploy: flake.lock cc-ci 2f6787f→a8b836e, lock commit 4e286ba pushed,
+/etc/cc-ci ff to a8b836e, detached nixos-rebuild test → health (0 failed, swarm 1/1) → switch
+(21:17 UTC). Bridge env confirms POLL_REPOS += recipe-maintainers/rustdesk-server (1/1, no auth
+errors). !testme posted on recipe PR #1 via testme-on-pr.sh → build 119 VERDICT=GREEN (level 5,
+lint pass, clean_teardown, no_secret_leak; card embedded in PR by bridge). Card saved:
+.cc-ci-logs/upgrades/rustdesk-server-testme-20261030.png. Next weekly /upgrade-all picks
+rustdesk-server up automatically (used-recipes.md row committed cd1ec01). /root/cc-ci-test-verify
+removed. Operator merges PR #1 then `abra recipe release`.

cc-ci-plan/REBOOTS.md

@@ -23,3 +23,4 @@ restarts the loops on boot. Count the lines below to see how often it's happenin
 - 2026-05-30 17:03:05 BST — reboot detected; loops auto-started by systemd (resuming phase index 6). boot_id=f565f752-0463-42db-b787-9e0db35a5e3f
 - 2026-05-31 03:38:29 UTC — reboot detected; loops auto-started by systemd (resuming phase index 5). boot_id=51c17fc3-8391-4109-bce2-413fbee6f26d
 - 2026-08-23 03:11:24 UTC — reboot detected; loops auto-started by systemd (resuming phase index 14). boot_id=8215fc5e-8032-4844-b2b8-e0f8c0d9b13e
+- 2026-09-28 19:54:59 UTC — reboot detected; loops auto-started by systemd (resuming phase index 14). boot_id=66dadc4e-7f44-493a-aee9-fa76952d78c3

cc-ci-plan/advisory-scan.py

@@ -231,6 +231,8 @@ def registry_urls(recipe: str, registry_dir: str) -> tuple[list[str], str | None
         # recipe render '?' for no real reason — that is what put immich and n8n in the unknown
         # column on 2026-08-07 (https://docs.n8n.io/release-notes/` ← note the backtick).
         u = u.rstrip("`'\"*.,;:>)")
+        if "…" in u or "..." in u:
+            continue  # prose placeholder, not a fetchable security source
         if u and u not in urls:
             urls.append(u)
     return urls, path
@@ -826,6 +828,7 @@ def scan(recipe: str, v_from: str | None, v_to: str | None, registry_dir: str,
                     windows[src] = (v_from, v_to)
                     window_key[src] = k
     for key, wf, wt in (images or []):
+        matched = False
         for src in gh_sources:
             if src in windows:
                 continue
@@ -837,6 +840,9 @@ def scan(recipe: str, v_from: str | None, v_to: str | None, registry_dir: str,
             if k in src.lower() or name in k:
                 windows[src] = (wf, wt)
                 window_key[src] = key
+                matched = True
+        if not matched:
+            report.setdefault("unmatched_image_windows", []).append(key)
     report["windows"] = {k: {"from": f, "to": t} for k, (f, t) in windows.items()}
 
     def _classify_window(src, wf, wt):
@@ -1020,24 +1026,22 @@ def scan(recipe: str, v_from: str | None, v_to: str | None, registry_dir: str,
     # A scan with NO usable source has not measured anything, so it must not report a number —
     # least of all 0, which asserts safety. mumble had no cc-ci-plan/upstream/mumble.md at all and
     # still produced "0 identified", which was then published as a clean 0 in a CVE report.
-    usable_sources = [
-        s for s in report["sources"]
-        if s["status"] == "ok" or s["status"].startswith("no-advisories-published")
-    ]
+    usable_sources = [s for s in report["sources"] if s["status"] == "ok"]
     no_sources = not usable_sources
     if no_sources:
         report["no_usable_sources"] = True
-    report["count_known"] = not unresolved_any and not no_sources
-    report["cve_count_fixed"] = (len(fixed_set)
-                                 if (not unresolved_any and not no_sources) else None)
-    report["cve_count_total_seen"] = len(report["cves"])
-    # Only GENUINE failures make a count unreliable. "no-advisories-published" (404: the repo has
-    # no advisory feed) and "skipped: template URL" are benign and must not degrade the verdict.
     report["sources_failed"] = [
         s["source"]
         for s in report["sources"]
         if not (s["status"] == "ok" or s["status"].startswith(("no-advisories-published", "skipped:")))
     ]
+    coverage_failed = bool(report.get("unmatched_image_windows")) or bool(v_from and not primary)
+    report["count_known"] = not unresolved_any and not no_sources and not coverage_failed and not report["sources_failed"]
+    report["cve_count_fixed"] = (len(fixed_set)
+                                 if report["count_known"] else None)
+    report["cve_count_total_seen"] = len(report["cves"])
+    # Only GENUINE failures make a count unreliable. "no-advisories-published" (404: the repo has
+    # no advisory feed) and "skipped: template URL" are benign and must not degrade the verdict.
     report["sources_benign"] = [
         s["source"]
         for s in report["sources"]
@@ -1052,16 +1056,20 @@ def markdown(rep: dict) -> str:
          f"{rep.get('from') or '?'} → {rep.get('to') or '?'}"]
     if not rep.get("count_known", True):
         L.append("\n**CVEs fixed by this upgrade: UNKNOWN — the scan could NOT determine a count.**")
+        if rep.get("fixed_by_this_upgrade"):
+            L.append(f"\nAt least {len(rep['fixed_by_this_upgrade'])} fixed CVE(s) were identified: "
+                     + ", ".join(rep["fixed_by_this_upgrade"][:15]))
         if rep.get("no_usable_sources"):
             L.append("\n⚠ This is NOT zero. **No usable source was checked at all** — the registry "
                      "file `cc-ci-plan/upstream/<recipe>.md` is missing or every source failed, so "
                      "nothing was measured. Render this recipe's cve cell as `?`, never `0`, and add "
                      "the registry file.")
         else:
-            L.append("\n⚠ This is NOT zero. A version-scheme change (e.g. semver → calver) makes "
-                     "numeric ordering meaningless across this jump, so no advisory could be "
-                     "classified. Render this recipe's cve cell as `?`, never `0`. Read the vendor's "
-                     "release notes for the jump and count by hand.")
+            L.append("\n⚠ This is NOT zero. At least one version window or advisory source could "
+                     "not be checked completely. Render this recipe's cve cell as `?`, never `0`.")
+        if rep.get("unmatched_image_windows"):
+            L.append("\n⚠ image windows with no matching advisory source: "
+                     + ", ".join(rep["unmatched_image_windows"]))
         if rep["unclassified"]:
             L.append(f"\nAdvisories seen but unclassifiable ({len(rep['unclassified'])}) — includes "
                      f"other images in this recipe: " + ", ".join(rep["unclassified"][:12]))

cc-ci-plan/audit-sources.py

@@ -234,6 +234,8 @@ def _registry_urls(recipe: str):
     urls = []
     for u in re.findall(r"https?://[^\s)|\]]+", text):
         u = u.rstrip("`'\"*.,;:>)")
+        if "…" in u or "..." in u:
+            continue
         if u and u not in urls:
             urls.append(u)
     return urls, path
@@ -272,21 +274,23 @@ def main() -> int:
         blind_recipes, noisy = [], 0
         for r in recipes:
             rows = security_source_audit(r)
-            feeds = [x for x in rows if x["kind"] == "advisory-feed" and x["cves"] > 0]
-            logs = [x for x in rows if x["kind"] == "changelog"]
+            feeds = [x for x in rows if x["kind"] == "advisory-feed" and x["cves"] > 0
+                     and x["status"] == "ok"]
+            logs = [x for x in rows if x["kind"] == "changelog" and x["status"] == "ok"]
             unattr = [x for x in rows if x["kind"] == "unattributable"]
+            failed = [x for x in rows if x["status"].startswith("error:")]
             noisy += len(unattr)
             usable = len(feeds) + len(logs)
             if usable == 0:
                 blind_recipes.append(r)
-                print(f"!! {r}: NO USABLE CVE SOURCE — {len(unattr)} unparseable page(s), "
-                      f"0 advisory feeds, 0 attributable changelogs")
+                print(f"!! {r}: NO VERIFIED CVE SOURCE — {len(unattr)} unparseable page(s), "
+                      f"0 populated advisory feeds, 0 attributable changelogs, {len(failed)} failed source(s)")
                 for x in rows:
-                    print(f"      {x['kind']:15} {x['source'][:64]} ({x['cves']} CVEs)")
+                    print(f"      {x['kind']:15} {x['source'][:64]} ({x['cves']} CVEs; {x['status']})")
             else:
                 print(f"OK {r}: {len(feeds)} advisory-feed(s), {len(logs)} changelog(s)"
-                      + (f", {len(unattr)} unparseable page(s) (redundant — covered by a feed)"
-                         if unattr else ""))
+                      + (f", {len(unattr)} unparseable page(s)" if unattr else "")
+                      + (f", {len(failed)} failed source(s) — counts incomplete" if failed else ""))
                 for x in logs:
                     print(f"      changelog  {x['source'][:62]} ({x['usable']}/{x['cves']})")
         print(f"\n{len(recipes)} recipes · {len(blind_recipes)} with NO usable CVE source"

cc-ci-plan/resolve-images.py

@@ -163,7 +163,8 @@ def parse_ref(ref: str) -> dict:
     return {"registry": host, "repo": repo, "tag": tag, "digest": digest}
 
 
-HUB_RECENT_PAGES = int(os.environ.get("RESOLVE_IMAGES_HUB_PAGES", "10"))
+HUB_RECENT_PAGES = int(os.environ.get("RESOLVE_IMAGES_HUB_PAGES", "40"))
+_HUB_TRUNCATED: set[str] = set()
 
 
 def _hub_tag_exists(repo: str, tag: str) -> bool:
@@ -177,10 +178,10 @@ def _hub_tag_exists(repo: str, tag: str) -> bool:
 def _hub_tags(repo: str) -> list[str]:
     """Recently-pushed tags, newest first.
 
-    Popular Docker Hub repos carry many thousands of tags, so a full enumeration is impractical —
-    but it is also unnecessary: a tag NEWER than the one we run must have been pushed AFTER it, so
-    ordering by last_updated and reading a bounded recent window is sufficient to find any upgrade.
-    (ghcr offers no ordering, which is why that path needs a different strategy.)
+    Popular Docker Hub repos carry many thousands of tags, so enumerate a bounded recent window.
+    A rebuilt old tag can displace a newer version from that window, so callers MUST treat a
+    truncated listing with no detected upgrade as unverified, not current. (ghcr offers no recency
+    ordering, which is why that path needs a different strategy.)
     """
     tags, url = [], (f"https://hub.docker.com/v2/repositories/{repo}/tags"
                      f"?page_size=100&ordering=last_updated")
@@ -191,6 +192,8 @@ def _hub_tags(repo: str) -> list[str]:
         url = d.get("next")
         if not url:
             break
+    if url:
+        _HUB_TRUNCATED.add(repo)
     return tags
 
 
@@ -236,15 +239,26 @@ def _oci_tags(host: str, repo: str) -> list[str]:
     hdrs = _oci_bearer(host, repo)
     tags, url = [], f"https://{host}/v2/{repo}/tags/list?n=1000"
     for _ in range(MAX_TAG_PAGES):
-        req = urllib.request.Request(url, headers={"User-Agent": UA, **hdrs})
-        with urllib.request.urlopen(req, timeout=TIMEOUT) as r:
-            tags += (json.load(r) or {}).get("tags") or []
-            link = r.headers.get("Link", "") or ""
+        for attempt in range(RETRIES):
+            req = urllib.request.Request(url, headers={"User-Agent": UA, **hdrs})
+            try:
+                with urllib.request.urlopen(req, timeout=TIMEOUT) as r:
+                    tags += (json.load(r) or {}).get("tags") or []
+                    link = r.headers.get("Link", "") or ""
+                break
+            except urllib.error.HTTPError as e:
+                if e.code not in (429, 503) or attempt == RETRIES - 1:
+                    raise
+                retry_after = e.headers.get("Retry-After") or ""
+                delay = min(int(retry_after) if retry_after.isdigit() else 2 ** (attempt + 1), 60)
+                time.sleep(delay)
         m = re.search(r'<([^>]+)>;\s*rel="next"', link)
         if not m:
             break
         nxt = m.group(1)
         url = f"https://{host}{nxt}" if nxt.startswith("/") else nxt
+    else:
+        raise RuntimeError(f"tag listing truncated after {MAX_TAG_PAGES} pages")
     return tags
 
 
@@ -299,6 +313,39 @@ def _release_fallback_repos(registry: str, repo: str) -> list[tuple[str, str]]:
     return cands
 
 
+def _oci_tag_exists(host: str, repo: str, tag: str) -> bool:
+    """Verify a release candidate is actually published as an image manifest."""
+    hdrs = {"User-Agent": UA,
+            "Accept": "application/vnd.oci.image.index.v1+json, "
+                      "application/vnd.docker.distribution.manifest.list.v2+json, "
+                      "application/vnd.oci.image.manifest.v1+json",
+            **_oci_bearer(host, repo)}
+    url = f"https://{host}/v2/{repo}/manifests/{urllib.parse.quote(tag, safe='')}"
+    try:
+        with urllib.request.urlopen(urllib.request.Request(url, headers=hdrs, method="HEAD"),
+                                    timeout=TIMEOUT) as response:
+            return response.status == 200
+    except urllib.error.HTTPError as e:
+        if e.code == 404:
+            return False
+        raise
+
+
+def _release_manifest_fallback(registry: str, repo: str, current: str) -> tuple[list[str], str] | None:
+    """For huge GHCR registries, use release tags only after checking actual manifests."""
+    for owner, name in _release_fallback_repos(registry, repo):
+        releases = github_release_tags(owner, name)
+        if current not in releases:
+            continue  # incomplete release window; it cannot establish a current baseline
+        candidates = sorted((tag for tag in set(releases)
+                             if shape(tag) == shape(current) and vkey(tag) >= vkey(current)),
+                            key=vkey, reverse=True)
+        for tag in candidates:
+            if _oci_tag_exists(registry, repo, tag):
+                return [current, tag], f"github-releases+registry-manifest:{owner}/{name}"
+    return None
+
+
 _TAG_CACHE: dict[tuple[str, str], tuple[list[str], str | None]] = {}
 
 
@@ -335,8 +382,16 @@ def resolve(ref: str) -> dict:
            "upgrade_available": False}
     tags, err = list_tags(info["registry"], info["repo"])
     if err:
-        out["status"] = f"error: {err}"
-        return out
+        try:
+            fallback = (_release_manifest_fallback(info["registry"], info["repo"], info["tag"])
+                        if "tag listing truncated" in err else None)
+        except Exception as e:  # noqa: BLE001 — a failed fallback must stay unresolved
+            out["status"] = f"error: registry {err}; release fallback {type(e).__name__}: {e}"
+            return out
+        if not fallback:
+            out["status"] = f"error: {err}"
+            return out
+        tags, out["source"] = fallback
     out["tags_seen"] = len(set(tags))
     # INTEGRITY CHECK: the tag we are currently running MUST appear in the listing. If it does not,
     # the listing is incomplete and any "newest" derived from it is a guess — ghcr paginates to tens
@@ -367,8 +422,9 @@ def resolve(ref: str) -> dict:
     same = [t for t in set(tags) if shape(t) == want and vkey(t)]
     out["candidates"] = len(same)
     if not same:
-        # Not a failure: digest-only pins and `latest`/`stable` have no comparable siblings.
-        out["status"] = "no comparable tags (shape has no numeric siblings)"
+        # A floating or unusual tag cannot prove that an image is current. Keep it visible to
+        # callers rather than letting an empty candidate set become a false UPTODATE verdict.
+        out["status"] = "unverifiable: no comparable numeric tags"
         return out
     newest = max(same, key=vkey)
     out["newest_same_shape"] = newest
@@ -377,6 +433,10 @@ def resolve(ref: str) -> dict:
         if within:
             out["newest_within_major"] = max(within, key=vkey)
     out["upgrade_available"] = bool(cur and vkey(newest) > cur)
+    if (info["registry"] in ("docker.io", "registry-1.docker.io")
+            and info["repo"] in _HUB_TRUNCATED and not out["upgrade_available"]):
+        out["status"] = (f"unverifiable: Docker Hub listing truncated after "
+                         f"{HUB_RECENT_PAGES * 100} recent tags")
     return out
 
 
@@ -436,16 +496,16 @@ def main() -> int:
         "recipe": a.recipe,
         "images": results,
         "upgrades_available": [r["ref"] for r in results if r["upgrade_available"]],
-        "unresolved": [r["ref"] for r in results if r["status"].startswith("error")],
+        "unresolved": [r["ref"] for r in results if r["status"].startswith(("error", "skipped", "unverifiable"))],
         # The whole point: distinguish "checked, current" from "could not check".
-        "all_resolved": not any(r["status"].startswith("error") for r in results),
+        "all_resolved": bool(results) and all(r["status"] == "ok" for r in results),
     }
     if not a.table:
         print(json.dumps(report, indent=2))
         return 0
     print(f"{a.recipe} — {len(results)} images")
     for r in results:
-        flag = "UPGRADE" if r["upgrade_available"] else ("ERROR" if r["status"].startswith("error") else "current")
+        flag = "UPGRADE" if r["upgrade_available"] else ("UNKNOWN" if r["status"] != "ok" else "current")
         print(f"  [{flag:7}] {r['repo']}:{r['tag']}" + ("  (digest-pinned)" if r["digest"] else ""))
         print(f"            shape={r['shape']}  candidates={r['candidates']}"
               f"  newest_same_shape={r['newest_same_shape']}"

cc-ci-plan/test-advisory-scan.py

@@ -22,6 +22,7 @@ import io
 import json
 import os
 import pathlib
+import tempfile
 import sys
 import unittest
 import unittest.mock
@@ -153,6 +154,12 @@ class TestRegistryUrls(unittest.TestCase):
         p.write_text(text)
         return A.registry_urls("r", tmp)
 
+    def test_prose_ellipsis_is_not_a_security_source(self):
+        with tempfile.TemporaryDirectory() as directory:
+            urls, _ = self._write(directory,
+                                  "https://github.com/cryptpad/cryptpad and example `https://…` URLs")
+        self.assertEqual(urls, ["https://github.com/cryptpad/cryptpad"])
+
     def test_strips_trailing_markdown_punctuation(self):
         # Production defect: a captured backtick 404'd the fetch and rendered n8n/immich as '?'.
         import tempfile
@@ -290,12 +297,14 @@ class TestPerImageWindows(unittest.TestCase):
         self.assertEqual(rep["windows"]["github-advisories:discourse/discourse"],
                          {"from": "3.5.3", "to": "3.5.4"})
 
-    def test_unmatched_image_name_is_silently_ignored(self):
-        # Documents CURRENT behaviour: a typo'd name costs coverage without warning.
+    def test_unmatched_image_name_makes_count_unknown(self):
+        # A typo'd sidecar name must not silently turn its CVEs into a clean zero.
         rep = run_scan([self.APP], v_from="3.5.3", v_to="3.5.4",
                        images=[("nosuchimage", "1", "2")], urls=self.URLS[:1])
         self.assertEqual(list(rep["windows"]), ["github-advisories:discourse/discourse"])
-        self.assertTrue(rep["count_known"])
+        self.assertFalse(rep["count_known"])
+        self.assertEqual(rep["unmatched_image_windows"], ["nosuchimage"])
+        self.assertIsNone(rep["cve_count_fixed"])
 
 
 class TestSchemeChangeDateFallback(unittest.TestCase):
@@ -377,6 +386,13 @@ class TestCountGuarantees(unittest.TestCase):
                        v_from="1.0", v_to="1.1")
         self.assertIn("github-advisories:app/app", rep["sources_failed"])
         self.assertIn("FAILED", A.markdown(rep))
+        self.assertFalse(rep["count_known"])
+
+    def test_only_404_feed_is_not_evidence_of_clean_scan(self):
+        rep = run_scan([gh("app/app", [], status="no-advisories-published")],
+                       v_from="1.0", v_to="1.1")
+        self.assertTrue(rep["no_usable_sources"])
+        self.assertFalse(rep["count_known"])
 
     def test_no_window_given_classifies_nothing(self):
         rep = run_scan([gh("app/app", [adv("CVE-1", patched="1.1")])])

cc-ci-plan/test-resolve-images.py

@@ -0,0 +1,55 @@
+#!/usr/bin/env python3
+"""Offline regressions for the weekly image survey's false-current cases."""
+
+import importlib.util
+import pathlib
+import unittest
+from unittest import mock
+
+HERE = pathlib.Path(__file__).resolve().parent
+spec = importlib.util.spec_from_file_location("resolve_images", HERE / "resolve-images.py")
+resolver = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(resolver)
+
+
+class TestResolverCoverage(unittest.TestCase):
+    def test_cryptpad_calver_tag_is_found_independently_of_abra(self):
+        with mock.patch.object(resolver, "list_tags", return_value=(
+            ["version-2026.5.1", "version-2026.10.0", "opendesk-2026.08.04"], None)):
+            row = resolver.resolve("cryptpad/cryptpad:version-2026.5.1")
+        self.assertTrue(row["upgrade_available"])
+        self.assertEqual(row["newest_same_shape"], "version-2026.10.0")
+
+    def test_floating_and_templated_tags_cannot_prove_current(self):
+        with mock.patch.object(resolver, "list_tags", return_value=(["alpine", "1.2.3"], None)):
+            floating = resolver.resolve("atmoz/sftp:alpine")
+        templated = resolver.resolve("ghost:${IMAGE_VERSION}-alpine")
+        self.assertTrue(floating["status"].startswith("unverifiable:"))
+        self.assertTrue(templated["status"].startswith("skipped:"))
+
+    def test_truncated_registry_listing_cannot_prove_current(self):
+        with mock.patch.object(resolver, "list_tags", return_value=(["1.2.3"], None)):
+            with mock.patch.object(resolver, "_HUB_TRUNCATED", {"library/example"}):
+                row = resolver.resolve("example:1.2.3")
+        self.assertTrue(row["status"].startswith("unverifiable:"))
+
+    def test_huge_ghcr_listing_uses_only_published_release_manifests(self):
+        ref = "ghcr.io/immich-app/immich-machine-learning:v3.3.5"
+        with mock.patch.object(resolver, "list_tags", return_value=([], "RuntimeError: tag listing truncated")), \
+             mock.patch.object(resolver, "github_release_tags",
+                               return_value=["v3.3.6", "v3.3.5"]), \
+             mock.patch.object(resolver, "_oci_tag_exists", side_effect=lambda h, r, t: t == "v3.3.5"):
+            row = resolver.resolve(ref)
+        self.assertEqual(row["status"], "ok")
+        self.assertFalse(row["upgrade_available"])  # release exists; image does not yet
+        self.assertIn("registry-manifest", row["source"])
+        with mock.patch.object(resolver, "list_tags", return_value=([], "RuntimeError: tag listing truncated")), \
+             mock.patch.object(resolver, "github_release_tags",
+                               return_value=["v3.3.6", "v3.3.5"]), \
+             mock.patch.object(resolver, "_oci_tag_exists", return_value=True):
+            upgraded = resolver.resolve(ref)
+        self.assertTrue(upgraded["upgrade_available"])
+
+
+if __name__ == "__main__":
+    unittest.main()

cc-ci-plan/upstream/cryptpad.md

@@ -2,11 +2,26 @@
 
 | service | image | source repo | releases / changelog |
 |---------|-------|-------------|----------------------|
-| app     | cryptpad/cryptpad | https://github.com/cryptpad/cryptpad | https://github.com/cryptpad/cryptpad/releases |
+| app     | cryptpad/cryptpad | https://github.com/cryptpad/cryptpad | https://github.com/cryptpad/cryptpad/releases · https://hub.docker.com/r/cryptpad/cryptpad/tags |
 | web     | nginx             | https://github.com/nginx/nginx       | https://nginx.org/en/CHANGES |
 
 ## Standing notes
 - `abra recipe upgrade` cannot parse the `version-YYYY.M.D` tag format of the cryptpad/cryptpad image,
   so the app image is bumped by hand; the nginx sidecar is what `abra recipe upgrade` actually moves.
+- Only use tags in the canonical `version-YYYY.M.x` scheme. The `opendesk-YYYY.MM.DD` tags on
+  Docker Hub are openDesk-specific builds, not newer CryptPad releases — e.g. `opendesk-2026.08.04`
+  (built 2026-08-04) still contains CryptPad `2026.5.1` in `/cryptpad/package.json`.
+- CryptPad's GitHub releases and Docker Hub tags are not simultaneous: `2026.10.0` was released
+  2026-10-21; `version-2026.10.0` appeared on Docker Hub on 2026-10-23 (first seen by the weekly
+  upgrade run on 2026-10-30). Re-check the Docker Hub tags page before treating a GitHub release as
+  available for the recipe.
+- The 2026.10.0 release is a breaking one for the recipe: full server rewrite (separate
+  `cryptpad-server` package), split `config.js` → `config/infra.js` (public origin/ports move),
+  one proxy port for HTTP + websocket (no separate `3003` websocket port), SSO plugin v1.0.0
+  required, and Euro-Office 9.3 replacing the OnlyOffice editors. Recipe handling in the
+  `upgrade-2026.10.0` PR: `CPAD_MAIN_DOMAIN`/`CPAD_SANDBOX_DOMAIN` must be full `https://…` URLs
+  (the new server assigns them verbatim to `infra.public.origin`/`sandboxOrigin`; the image's
+  entrypoint auto-creates `config/infra.js`), nginx `/cryptpad_websocket` proxies to `:3000`,
+  `abra.sh` config/nginx/SSO-entrypoint version labels bumped, SSO default 1.0.0.
 - nginx is HTTP/1.1-only here (sidecar on :8083, `proxy_http_version 1.1`), so HTTP/2/3 CVE changes in
   nginx releases generally don't affect this recipe — but still note them.

cc-ci-plan/upstream/custom-html.md

@@ -7,6 +7,7 @@
 
 ## Standing notes
 - nginx even-numbered minor versions (1.28.x, 1.30.x) are mainline; odd (1.27.x, 1.29.x) are stable. The recipe tracks mainline.
+- nginx 1.31.7 (14 Oct 2026) is a security patch release with FIVE security fixes: CVE-2026-15176 (buffer overflow in a worker using `alias` when the URI was changed by `rewrite`), CVE-2026-93479 (buffer overflow evaluating string expressions with variables — affects document-root calculation and request logging in http/stream), CVE-2026-95101 (HTTP/3: part of an unknown frame treated as request body — bypasses `client_max_body_size` / injects into proxied request), CVE-2026-97629 (TLS session-ticket key rotation left one key uninitialized — session forgery), CVE-2026-97205 (use-after-free with async file I/O when an SSI page has `include virtual`). Features: geoip MMDB support, backend SSL cert verification by IP, QUIC reconfiguration + binary upgrade with `quic_bpf`. Change: `ngx_http_v2_module` now uses RFC 9218 prioritisation (RFC 7540 priority signals ignored). Bugfixes (X-Accel-Redirect cached response segfault, dav COPY/MOVE with alias segfault, QUIC flood false positives, SSL close_notify handling). NONE affect the custom-html static-file-serving use case (no HTTP/3/QUIC, no AIO/SSI, no `rewrite`+`alias` path mapping, no session-ticket rotation config) — but the 5 CVEs are why this bump is warranted. No breaking config changes; no operator action beyond the image pull.
 - compose.git-pull.yml is an optional overlay for git-pull functionality; alpine/git version lives there.
 - compose.sftp.yml and compose.sso.yml are other optional overlays; linuxserver/openssh-server uses `latest` tag and is not version-pinned.
 - nginx 1.31.6 (15 Sep 2026) is a security patch release: CVE-2026-90439 (heap buffer overflow in a worker under certain configurations when using HTTP/3 with OpenSSL 3.5.0 and earlier). Change: QUIC transport parameters extension received in an SSL connection is now always ignored. Bugfixes: binary upgrade with control API socket + perl module; predicate-location evaluation error ignored; nested location lookup with regex/predicate locations; segfault reading config with `geo` … `ranges` + corrupted binary base file. None affect the custom-html static-file-serving use case (no HTTP/3/QUIC, no predicate locations, no perl, no `geo`).

cc-ci-plan/upstream/gitea.md

@@ -41,3 +41,7 @@ in this file, so keep vendor security pages listed here.
 - **1.26.3**: carries a regression (#38177 "context deadline exceeded" opening repo code pages) — upstream says upgrade straight to 1.26.4, skip 1.26.3.
 - **1.27.0** (released 2026-07-13, MINOR with BREAKING changes): `Feat(actions)!: improve support for reusable workflows` (#37478) and `Use Content-Security-Policy: script nonce` (#37232, may break custom inline JS / reverse-proxy CSP). Many security + feature additions. A dedicated minor-bump run should evaluate CSP / reusable-workflow impact before adopting.
 - **1.27.1** (released 2026-07-27, PATCH): security fix (oauth2 mandatory 2FA enforcement on authorize/grant endpoints, #38606), API swagger alignment, diff-contrast enhancement, and many bugfixes (actions reusable-workflow/job-stranding, OIDC end-session, repo-deletion cleanup, webhook/mail). No breaking changes; safe patch bump from 1.27.0. Release notes: https://github.com/go-gitea/gitea/releases/tag/v1.27.1
+- **28.0.0** (released 2026-09-29, NEW MAJOR, BREAKING): git network operations now route through an internal proxy and egress settings changed (`Fix(git)!`, #39426) — outbound Git may need `[egress]` allow-listing; Actions adds `RUN_RETENTION_DAYS` (#38855). Also a large SECURITY batch (SSH key fingerprint identification, rejected invalid/duplicate Git objects, repo-scoped team authz, x/crypto DoS). GitHub notes: https://github.com/go-gitea/gitea/releases/tag/v28.0.0 — vendor post: https://blog.gitea.com/release-of-28.0.0/
+- **28.1.0** (released 2026-10-06, MINOR): 9 security fixes (push-mirror permission/`DISABLE_NEW_PUSH`, local-path import authz, repository media stored XSS, OAuth2 refresh-token kind, issue-attachment authz, `FORCE_PRIVATE` push-to-create, private RSS/Atom feed leak, tag-delete deleting releases, migration page-size DoS), plus MySQL/MariaDB `READ COMMITTED` and go-git repack fixes. No breaking changes. GitHub notes: https://github.com/go-gitea/gitea/releases/tag/v28.1.0 — vendor post (CVE list): https://blog.gitea.com/release-of-28.1.0/
+- **28.2.0** (released 2026-10-18, MINOR): 9 newly-fixed Gitea CVEs (SSH cert principal confusion, OpenID linking `state`, locked-issue reactions authz, release-attachment metadata leak, `/{username}/-/repositories` panic, draft-release attachment disclosure, Actions artifact v4 body limit, CSRF on repo topics, cross-org label deletion) plus dependency CVE-2026-78659 (`golang.org/x/net` v0.60.0). Three more CVEs are disclosed here but were already fixed in ≤28.1 (CVE-2026-110324, CVE-2026-109844, CVE-2026-97859). No breaking changes and no Postgres requirement. GitHub notes: https://github.com/go-gitea/gitea/releases/tag/v28.2.0 — vendor post (CVE list): https://blog.gitea.com/release-of-28.2.0/
+- **app.ini writability**: Gitea 1.24+ writes runtime config (e.g. `WORK_PATH`) back to `app.ini`; mounting it directly from a read-only swarm config makes that write fail (non-fatal but warned). The unmerged mirror PR #4 (`fix-app-ini-writable`) seeds a writable copy in the `config` volume. Image bumps must not change the config mount / `docker-setup.sh` handling on their own.

cc-ci-plan/upstream/immich.md

@@ -95,8 +95,45 @@
   (v3.2.2→v3.2.4, app+ML only, fresh PR). Direct registry check confirmed
   `immich-server/ml:v3.2.4` exist; `abra recipe upgrade` still FATAs on the tag+digest DB pin.
   Valkey CVE note: CVE-2026-56684 / CVE-2026-63639 (UAF→RCE) are fixed in 9.1.1 and are therefore
-  **already closed** by the recipe's pinned 9.1.1 — not by this app bump; ignore the live-`9`-tag
+  **  already closed** by the recipe's pinned 9.1.1 — not by this app bump; ignore the live-`9`-tag
   drift. CVE-2026-25243 (valkey RESTORE zipmap) is vendor-page-only and undecided.
+- **immich-server v3.3.0 (2026-10-07) + v3.3.1 (2026-10-08, latest stable) pins** (from
+  `docker/docker-compose.yml` @ the v3.3.1 tag): `postgres:14-vectorchord0.4.3-pgvectors0.2.0@sha256:bcf63357…`
+  (UNCHANGED vs v3.2.4) and `valkey:9@sha256:c123e3715db63d06d4ad6964884037aa0d5d4d703939b9929954112889708e1d`
+  — the **valkey pin moved** 9.1.1 (`70739f85…`) → **9.1.2 (`c123e371…`)**, which is now immich's own
+  official tested pin (the live `9` tag has since drifted to `418652cf…`; keep immich's `c123e371…`
+  per the 2026-07-17 precedent). `example.env` @ v3.3.1 is **byte-identical** to v3.2.4. v3.3.0 is a
+  feature release (people sharing / cluster groups, birthday memories, auto-stacked mobile edits,
+  higher-quality thumbnails via linear-light resampling, OAuth claims synced on every login, PP-OCRv6
+  OCR models); v3.3.1 is a patch (people-sharing + mobile scroll fixes). **No breaking server-side
+  changes, no required config, no operator action.** New TypeORM migrations auto-run
+  (`ConvertUserOAuthIdEmptyStringToNull`, `RenameGeoNamesCountries`, `PersonSharing`,
+  `AddPersonUserTableSharedBySharedWithConstraint`) — all succeeded in a live v3.2.4→v3.3.1 deploy.
+  Opt-in only: `MACHINE_LEARNING_MODEL_REVISION=v2` (optimized models, numerically equivalent, no job
+  re-run) and the people-sharing feature (user-triggered button, not a deploy migration). Done in the
+  2026-10-09 upgrade (v3.2.4→v3.3.1, app+ML+valkey, extending PR #6; fresh `!testme` GREEN — build #45).
+- **immich-server v3.3.3 (2026-10-14) + v3.3.4 (2026-10-20) + v3.3.5 (2026-10-21, latest stable) pins** (from
+  `docker/docker-compose.yml` @ the v3.3.5 tag): `postgres:14-vectorchord0.4.3-pgvectors0.2.0@sha256:bcf63357…`
+  (UNCHANGED since v3.3.0) and `valkey:9@sha256:c123e371…` (immich's tested 9.1.2, also unchanged). Upstream
+  coopcloud published v3.3.4 as `1.12.0+v3.3.4` (`6b294fce`) and moved the recipe's `redis` pin to the then-live
+  `418652cf…`; that divergence is left as upstream has it (no reason to churn a sidecar in an app bump).
+  `docker/docker-compose.yml` and `docker/example.env` are byte-identical between v3.3.4 and v3.3.5.
+  v3.3.3 = bug fixes (mobile memories date, people-query perf); v3.3.4 = further people-query perf + translation;
+  **v3.3.5 = security patch** (certifi CA-trust fix + October 2026 Node.js security patches). No schema
+  migration, no config/env change, no breaking server-side change. Done in the 2026-10-23 upgrade
+  (v3.3.4→v3.3.5, app+ML only, extending PR #6; live v3.3.4→v3.3.5 dev deploy healthy — migrations finished,
+  `vector`/`vchord` present, no pgvectors drop; `!testme` GREEN first run — build #78).
+- **2026-10-30: postgres sidecar pgvectors 0.2.0 → 0.3.0 (same PG14 + VectorChord 0.4.3 scheme).** With
+  app/ML already at v3.3.5, the only newer same-scheme tag is
+  `ghcr.io/immich-app/postgres:14-vectorchord0.4.3-pgvectors0.3.0` (multi-arch OCI index digest
+  `sha256:87c050465fb969a68c7ac23e375e21f4c95cfacd0edce5fa1bc31e63b7891891`; re-verified via ghcr +
+  `docker buildx imagetools inspect`). immich v3.3.5's own compose still names 0.2.0, but immich only
+  supports `vchord`/`vector` now — `vectors` (pgvecto.rs) is not in its `VECTOR_EXTENSIONS`, so the
+  sidecar library is unused; 0.3.0 is backward-compatible (`ALTER EXTENSION vectors UPDATE`), immich
+  added 0.3.0 support in immich-app/immich#11624, and this recipe previously shipped 0.3.0 with
+  v3.0.1/v2.7.5. Direct dev deploy: 0.2.0-created volume reused by the 0.3.0 task, PG14 clean start,
+  `vchord 0.4.3`/`vector 0.8.1` installed, `/api/server/version` = 3.3.5, `!testme` GREEN (build #105).
+  Recipe PR extends upstream main `ac03d23`; publish with `abra recipe release immich -z`.
 - **2026-08-07 INFRA note: cc-ci runner's gitea clone-token is STALE (HTTP 401).** `!testme` build
   #1210 died at the recipe `git clone` step (~5s, before any deploy) with `could not read Username for
   'https://git.autonomic.zone'` — the runner's mounted gitea token `13e299f2…` is rejected (verified

cc-ci-plan/upstream/lasuite-docs.md

@@ -6,6 +6,9 @@
 | backend | lasuite/impress-backend | https://github.com/suitenumerique/docs | https://github.com/suitenumerique/docs/releases |
 | celery | lasuite/impress-backend | https://github.com/suitenumerique/docs | https://github.com/suitenumerique/docs/releases |
 | y-provider | lasuite/impress-y-provider | https://github.com/suitenumerique/docs | https://github.com/suitenumerique/docs/releases |
+| yhub | lasuite/impress-yhub | https://github.com/suitenumerique/docs | https://github.com/suitenumerique/docs/releases |
+| yhub-valkey | valkey/valkey | https://github.com/valkey-io/valkey | https://github.com/valkey-io/valkey/releases |
+| jwt-keys | lasuite/impress-yhub (Node crypto sidecar) | https://github.com/suitenumerique/docs | https://github.com/suitenumerique/docs/blob/main/UPGRADE.md |
 | docspec | ghcr.io/docspec/api | https://github.com/docspec/docspec | https://github.com/docspec/docspec/releases |
 | db | pgautoupgrade/pgautoupgrade | https://github.com/pgautoupgrade/docker-pgautoupgrade | https://github.com/pgautoupgrade/docker-pgautoupgrade/releases |
 | redis | redis | https://github.com/redis/redis | https://hub.docker.com/_/redis/tags |
@@ -13,8 +16,20 @@
 | web | nginx | https://github.com/nginx/nginx | https://nginx.org/en/CHANGES |
 
 ## Standing notes
-- impress-frontend, impress-backend (used for both backend + celery services), impress-y-provider all
-  share the same version tag from the suitenumerique/docs monorepo. Upgrade all four in lockstep.
+- **v6.0.x replaces the collaboration server**: hocuspocus/y-provider collaboration is gone; a new
+  `yhub` service (port 3002) serves `/collaboration/*`, with its own `yhub` PostgreSQL database and a
+  Valkey store. `y-provider` is conversion-only. New RSA keys (`JWT_PRIVATE_KEY_FILE` backend,
+  `YHUB_JWT_PRIVATE_KEY_FILE` yhub), `COLLABORATION_SERVER_SECRET` removed. Docs must be migrated into
+  yhub (`SOFT_MIGRATION=true` + `backend migrate_documents`). See UPGRADE.md `[6.0.0]`.
+- **v6.0.2** moves yhub to `@y/hub 0.10.1`, adding the `yhub_ydoc_versions_v1` table/indexes: re-run the
+  yhub schema script (`init-db`) — the recipe does it on every yhub start.
+- **The recipe generates the two RSA keys at runtime** into a `jwt_keys` volume (`jwt-keys` sidecar)
+  because abra cannot generate PEM secrets (`generate=false`) and Swarm secrets must pre-exist.
+- **docspec**: since Docs v5.7.0 the Rust `ghcr.io/docspec/api` (port 3000) replaces the archived Elixir
+  `ghcr.io/docspecio/api` (port 4000); Docs v6.0.2 expects `DOCSPEC_API_URL=http://docspec:3000/conversion`.
+- impress-frontend, impress-backend (used for both backend + celery services), impress-y-provider,
+  impress-yhub (and the jwt-keys sidecar image) all share the same version tag from the
+  suitenumerique/docs monorepo. Upgrade all of them in lockstep.
 - AUTO_MIGRATIONS=true means DB migrations run automatically on backend startup. No manual step needed.
 - Minio tag uses a date-based RELEASE.YYYY-MM-DDTHH-MM-SSZ format — abra cannot parse it for upgrades;
   check manually on https://github.com/minio/minio/releases.

cc-ci-plan/upstream/n8n.md

@@ -302,3 +302,87 @@
   `Array.merge`→`mergeIntoObject`; 2.37.0 JSON content-type on decorator body routes + binary-data
   endpoint adapt; 2.39.0 workflow-version endpoint deprecation; 2.40.x `POST /rest/login`
   `emailOrLdapLoginId` rename). Recommended release: `-y`.
+- 2.42.3 (2026-10-05, patch): API fix (let global members grant variable scopes to API keys) + core
+  fix (keep the task runner running on an unhandled promise rejection) + editor feature (Gateway
+  credits promotions on community nodes).
+- 2.42.4 (2026-10-07, patch): release body empty (republish / no listed changes).
+- 2.42.5 (2026-10-08, patch): 1 editor fix (show the Agent artifact while the builder creates it).
+- 2.42.6 (2026-10-09, patch): core feature (block create/update of workflows with deprecated nodes) +
+  Gemini node feature (Nano Banana 2.1 + 2 Lite in the image-model list). Backport of the 2.43.3 pair
+  onto the stable 2.42 line.
+- 2.42.7 (2026-10-12, patch): 2 core fixes (concurrent credential tests no longer share one node-type
+  registry; resume tool calls for deferred tools a skill loaded).
+- 2.43.0 (2026-10-06, Pre-release; the 2.43 feature minor): a large feature/bugfix release. Features:
+  Agent Builder / Instance AI / editor improvements + ~many core/API/editor fixes (valid schemas for
+  untyped values in the Public API spec; global members can grant variable scopes to API keys; return
+  an execution when its stored trace context is incomplete; Azure OpenAI Chat Model; etc.). **No
+  breaking compose/env/migration changes; no `N8N_*` env renames.** (2.43.1/2.43.2 = patch bugfixes:
+  editor Assistant/Execute UI; Kimi K3 + OpenAI-compatible Assistant providers self-hosted; Agent
+  artifact preview; agent-builder prompt-caching/background-sandbox perf.)
+- 2.43.3 (2026-10-09, Pre-release): core feature (block create/update of workflows with deprecated
+  nodes) + Gemini node feature (Nano Banana 2.1 + 2 Lite in the image-model list).
+- 2.43.4 (2026-10-12, patch, **stable line**): 2 core fixes (handle each Chat Hub segment once in
+  queue mode; resume tool calls for deferred tools a skill loaded).
+- 2.43.5 (2026-10-14, patch, stable line): core fix (validate the `size` arg of the `chunk()`
+  expression extension) + editor feature (show Gemini models by name / prioritised first).
+- 2.43.6 (2026-10-15, patch, **stable line — `stable` floating Docker tag points here**, digest
+  `sha256:dd580fe1…`): 1 core fix (create an XML parser per webhook request body). Backport of the
+  2.44.2 fix onto the stable 2.43 line.
+- 2.44.0 (2026-10-13, **Pre-release / beta line — `beta` floating Docker tag points to 2.44.2**): the
+  2.44 feature minor. Large core/editor change set (credential/node-type policy enforcement across
+  agent tools, credential tests/probes, package import, folder/credential/delete-transfer moves;
+  block the assistant when AI credits hit zero; per-process crash-journal paths; configurable
+  healthcheck timeout; Microsoft SharePoint Trigger node; many bugfixes). **No breaking
+  compose/env/volume/label changes; no `N8N_*` renames.**
+- 2.44.1 (2026-10-14, Pre-release/beta): 1 editor feature (Gemini models by name / prioritised first).
+- 2.44.2 (2026-10-15, Pre-release/beta; **newest numeric tag abra lists**; Docker Hub `beta` floating
+  tag = `sha256:b57e7a9c…`): 1 core fix (create an XML parser per webhook request body).
+- 2026-10-09 run (not separately logged here): PR #9 (branch `upgrade-a144baf`) extended from the
+  2.42.2 of the 2026-10-02 run to **2.43.2** (newest numeric tag abra listed that day). 2.43.2 was
+  still Pre-release on GitHub at the time (tracking-the-newest-tag precedent).
+- 2026-10-16 run: reconcile confirmed coopcloud upstream main still `9141e23` (3.5.1+2.40.5) and
+  PR #9 (branch `upgrade-a144baf`, at 2.43.2) still open/unmerged, so per the one-evolving-PR
+  contract the run extends PR #9 **2.43.2 → 2.43.6**. Operator guidance this run: verify whether
+  2.44.x is the newest STABLE line and take the newest SAFE patch. **2.44.x is the beta/pre-release
+  line** — GitHub marks 2.44.0/2.44.1/2.44.2 `prerelease=true`, and Docker Hub's `beta` floating tag
+  points at 2.44.2 while `stable` points at 2.43.6 (verified live). So the safe target is **2.43.6**
+  (newest patch on the now-stable 2.43 line), NOT the beta 2.44.2. 2.43.2→2.43.6 is patch-only
+  (2.43.3 deprecated-node workflow guard + Gemini image list; 2.43.4 Chat-Hub queue-mode + deferred
+  tools; 2.43.5 `chunk()` size-arg validation + Gemini picker; 2.43.6 per-request XML parser). No
+  breaking compose/env/migration changes. Live-deployed on cc-ci `dev-n8n` (2.40.5 → 2.43.6): TypeORM
+  migrations completed cleanly, editor `/rest/login` 401 (auth-gated, as expected), `/healthz` 200 —
+  converged. Advisory scan 2.43.2→2.43.6: 0 CVEs fixed (9 sources, none failed). Recommended
+  release: `-y` (relative to the published label app tag 2.40.5 the delta spans feature minors
+  2.41/2.42/2.43).
+- 2.43.7 (2026-10-16, patch): core suspend in-flight production executions at worker shutdown.
+- 2.43.8 (2026-10-19, patch): core turn off expression spans by default (backport of 2.44.4).
+- 2.44.3 (2026-10-16, patch): core keep Observer detail lines instead of discarding the response.
+- 2.44.4 (2026-10-19, patch): core reuse connections in the log-streaming webhook destination; turn
+  off expression spans by default.
+- 2.44.5 (2026-10-21, patch): core key queue job outcomes by job on shutdown resume; editor load
+  credential types on startup.
+- 2.44.6 (2026-10-22, patch; **`stable` floating tag points here**): Postgres Node prevent global
+  pgtypes overwrite.
+- 2.45.0 (2026-10-20, Pre-release; the 2.45 feature minor): automated migrations for removed nodes;
+  node and credential type policies enabled by default; durable scheduler enabled by default;
+  Idempotency-Key on Public API POSTs; workflow migration-findings endpoints/notifications; Kafka v2
+  default for new nodes; Monday.com V2 default; AI Agent/Assistant, editor and node fixes; UI rebrand.
+  **No breaking compose/env/volume/label changes; no `N8N_*` renames.**
+- 2.45.1 (2026-10-21, Pre-release): core claim an idle worker's next queue job and its lock atomically;
+  Microsoft Teams Trigger Graph subscription renewal via lifecycle notifications; editors can triage
+  migration findings.
+- 2.45.2 (2026-10-22, Pre-release; **`beta` floating tag points here — newest numeric tag**): core read
+  type-availability policies from memory at enforcement points; Postgres Node prevent global pgtypes
+  overwrite; migration report CSV export; OAuth-confirmation/email logo update.
+- 2026-10-23 run: reconcile confirmed coopcloud upstream main moved to `c834df2` ("NODE_FUNCTION_ALLOW_EXTERNAL
+  in env file", on top of 3.5.1+2.40.5) and PR #9 (branch `upgrade-a144baf`, at 2.43.6) still
+  open/unmerged. Per the one-evolving-PR contract, PR #9 extended **2.40.5 → 2.45.2** (explicit target
+  = highest tag abra lists). 2.45.x is the current Pre-release/beta line; **stable = 2.44.6** (verified
+  live via Docker Hub floating tags), flagged to the operator. Window spans feature minors 2.41–2.45,
+  but release notes show **no breaking compose/env/volume/label changes and no `N8N_*` renames**.
+  Live-deployed on cc-ci `dev-n8n`: TypeORM migrations completed cleanly through
+  `CreatePromotionRunAndReviewTables1792139680417`, editor accessible, `/healthz` 200, `/rest/login`
+  401; runtime deprecation warnings only (`WEBHOOK_URL`→`N8N_WEBHOOK_URL`, internal runner mode,
+  unverified-packages default, runner-timeout/compression-limit default reductions). Advisory scan
+  2.40.5→2.45.2: 0 CVEs identified, BUT the GitHub advisory sources failed (unauthenticated rate
+  limit) — scan incomplete, not to be read as clean. Recommended release: `-y`.

cc-ci-plan/upstream/rustdesk-server.md

@@ -0,0 +1,23 @@
+# Upstream sources — rustdesk-server
+
+| service | image | source repo | releases / changelog |
+|---------|-------|-------------|----------------------|
+| app | rustdesk/rustdesk-server | https://github.com/rustdesk/rustdesk-server | https://github.com/rustdesk/rustdesk-server/releases |
+| signal | rustdesk/rustdesk-server | https://github.com/rustdesk/rustdesk-server | https://github.com/rustdesk/rustdesk-server/releases |
+
+## Standing notes
+- **NOT enrolled** in the cc-ci weekly set (operator 2026-10-30: single client, maintenance contract
+  pending). Mirror `recipe-maintainers/rustdesk-server` exists + this registry entry so one-off
+  `/recipe-upgrade` runs and advisory scans work; if the client confirms, enroll via `/recipe-enroll`
+  (test suite + bridge + used-recipes weekly row).
+- Both services (`app` = hbbs, `signal` = hbbr) use the SAME image — always bump them together.
+- Docker Hub has stray `1.4.5-*` arch tags (pushed 2026-01-12, top-level `1.4.5` tag deleted, no
+  GitHub release) — an aborted/alias push, NOT a real version line. The real line is `1.1.x`;
+  GitHub "Latest" + docker `latest` agree (1.1.16 as of 2026-07-20). Ignore 1.4.5.
+- Recipe uses `-k _` on both hbbs and hbbr (enforces key verification, enforced-key mode since
+  1.1.11) and publishes host-mode ports 21115-21119 (hbbs) + 21117/21119 (hbbr) — direct
+  client connections, only 21118/21119 are the web-socket/http endpoints behind nothing; the
+  traefik labels on `app` are vestigial (hbbs has no HTTP :80 service; healthcheck commented out).
+- 1.1.15 raised **default bandwidth limits** and added auth-failure logging (fail2ban-friendly);
+  1.1.16 fixed **unauthenticated UDP punch-hole reflection/amplification abuse** + critical `mio`
+  0.8.5→0.8.11 issue + i32 overflow (peers offline 24.9–49.7d shown online).

cc-ci-plan/used-recipes.md

@@ -28,10 +28,11 @@ lasuite-drive     weekly
 lasuite-meet      weekly
 mailu             weekly
 matrix-synapse    weekly
-mattermost-lts    weekly
+mattermost-lts    external   # maintained elsewhere — used/tested by cc-ci but NOT weekly-upgraded
 mumble            weekly
 n8n               weekly
 plausible         external   # maintained elsewhere — used/tested by cc-ci but NOT weekly-upgraded
+rustdesk-server   weekly
 uptime-kuma       external   # maintained elsewhere — used/tested by cc-ci but NOT weekly-upgraded
 wordpress         weekly
 ```

flake.lock

@@ -10,11 +10,11 @@
         ]
       },
       "locked": {
-        "lastModified": 1790010051,
-        "narHash": "sha256-0dz8fg4zqg54tk+1A7CHZAw+lK9vWxre6Tioi5B0ocU=",
+        "lastModified": 1793644559,
+        "narHash": "sha256-u6xPHneKwIrd1taLVs1X12dUs46RzTzkZ9fJ8d/ZvYI=",
         "ref": "refs/heads/main",
-        "rev": "2f6787f0433e7ff3053116039f10c00cde34128f",
-        "revCount": 1558,
+        "rev": "d569e6528f2eba4309db4794b091d6a1ecc029bf",
+        "revCount": 1589,
         "type": "git",
         "url": "https://git.autonomic.zone/recipe-maintainers/cc-ci.git"
       },

memory/MEMORY.md

@@ -1,5 +1,7 @@
 # Memory index
 
+- [Enroll gate fixtures must fail loudly](enroll-gate-fixtures-fail-loudly.md) — version-gate reads via lifecycle.deployed_identity, never None-into-if; post-host-merge deploy = --cc-ci-only after merge (staging obsolete)
+
 - [Orchestrator host: Hetzner](orchestrator-host-hetzner.md) — runs on Hetzner cpx22; rebuild cmd, loops-service bounce, git-identity gotcha
 - [Push commits to remote](push-commits-to-remote.md) — push to git.autonomic.zone right after every commit in this repo
 - [Regression canary cadence](regression-canary-cadence.md) — server E2E canaries run on polish/review/release, not every commit

memory/enroll-gate-fixtures-fail-loudly.md

@@ -0,0 +1,43 @@
+---
+tags: [cc-ci, enrollment, testing, deploy]
+---
+
+# Enroll-verify gate fixtures must fail loudly
+
+## Context
+
+During rustdesk-server enrollment (2026-10-30, cc-ci PR #48) the first full-suite verification
+run came back RED only because of the *fixture*, not the app: the punch-hole regression test's
+version gate read the deployed image with a hand-rolled `docker service inspect`, got None, and
+the `if version is not None and version < (1,1,16)` gate silently decided to RUN on a 1.1.12 head
+where pre-fix answering is the expected behavior. An expected-RED looked like a real bug.
+
+## Lesson
+
+- A test gate that cannot decide its condition must `pytest.fail` LOUDLY with the raw value —
+  never default to None-into-an-if (silently runs → wrong RED) or silent skip (vacuous green).
+- cc-ci already has a first-class image reader: `runner/harness/lifecycle.py::deployed_identity(domain)`
+  → `{"version","image","chaos"}`; parse the tag from `.image` and strip any `@sha256:` digest.
+  Hand-rolled service inspects in recipe_meta/custom tests are a duplication hazard.
+- Version-gated custom tests skip on pre-fix heads by design (STYLE §2) — that is fine as long
+  as the version read itself cannot fail silently.
+
+# cc-ci host deploy mechanism (post host-merge, 2026-09+)
+
+To run any merged cc-ci main change on the live host now, use the `--cc-ci-only` path from
+/cc-ci-orchestrator-update — merge the PR first (enrollment PRs: comment verification evidence,
+then merge — invocation = authorization), then:
+
+```
+ssh cc-ci 'sudo -H -u loops bash -lc "cd /srv/cc-ci-orch && nix flake update cc-ci &&
+  git -c user.name=auto-update -c user.email=auto-update@cc-ci commit -qm \"flake.lock: cc-ci -> main\" -- flake.lock &&
+  git push -q origin HEAD:main"'
+ssh cc-ci 'git -C /etc/cc-ci pull --ff-only --recurse-submodules'
+systemd-run --unit=ccci-rebuild-test --collect -E HOME=/root -p WorkingDirectory=/srv/cc-ci-orch \
+  bash -lc "nixos-rebuild test --flake /srv/cc-ci-orch#cc-ci"   # := health checks := then switch
+```
+
+The old staging procedure (branch clone at /root/cc-ci-deploy, referenced by the retired
+/cc-ci-server-update skill and old enroll SKILL.md step 7 lines) is obsolete — the combined-host
+flake takes the cc-ci module from the lockfile, so a BRANCH cannot be staged that way; deploy
+after merge instead.