db-storage-bloat
Audit storage operability — VACUUM / autovacuum health and table-and-index bloat, transaction-ID wraparound risk (sev-5), and tombstone accumulation on wide-column stores. Module M18. Feeds the Performance & Scale score.
How do I install this agent skill?
npx skills add https://github.com/hainrixz/claude-db --skill db-storage-bloatIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill is a database storage audit tool that monitors health metrics such as table bloat and transaction ID wraparound risk. While functionally legitimate and containing integrity guidelines, it processes external database configurations and metadata which serves as a potential surface for indirect prompt injection.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
db-storage-bloat (M18)
Storage operability keeps a database fast and alive over time; it is a Performance & Scale (axis
performance) concern, feeding relational Almacenamiento w12 (shared with M22/M20/M21) and the
Tombstones category in the wide-column profile. The catastrophic case — TXID wraparound — can
force a Postgres shutdown, so it is a severity:5 cap when imminent.
What it checks
- VACUUM / autovacuum health — autovacuum disabled or mis-tuned on a high-churn table, dead tuples accumulating, bloated tables/indexes inflating size and slowing scans.
- TXID wraparound —
age(datfrozenxid)(or per-tablerelfrozenxid) approachingautovacuum_freeze_max_age/ the 2-billion ceiling. Imminent wraparound isseverity:5,fail(perf) — but only with live age evidence; statically it is at most aneeds_apinudge. - Tombstones (wide-column) — Cassandra/Scylla heavy deletes / TTL churn producing tombstones that
degrade reads (
tombstone_warn_thresholdterritory).
Score / axis
Feeds performance only (relational Almacenamiento w12; Tombstones in the wide-column profile).
Tier-0 (static)
Almost all of M18 is runtime truth. Statically, detect autovacuum-disabling DDL/config
(autovacuum_enabled = false, aggressive fillfactor), high-delete/TTL access patterns, and
wide-column delete-heavy modeling. Dead-tuple counts, real bloat, frozen-xid age, and tombstone counts
all require a live DB → status: needs_api at Tier-0 (never a silent pass).
Tier-1/2 (verification query, Postgres)
-- TXID wraparound headroom (Tier-1):
SELECT datname, age(datfrozenxid) AS xid_age,
current_setting('autovacuum_freeze_max_age')::int AS freeze_max
FROM pg_database ORDER BY age(datfrozenxid) DESC;
-- dead tuples + last autovacuum (Tier-2):
SELECT relname, n_dead_tup, n_live_tup, last_autovacuum
FROM pg_stat_user_tables ORDER BY n_dead_tup DESC LIMIT 20;
Method query_stat. xid_age near freeze_max (or the 2e9 ceiling) confirms wraparound as
established and capping. High n_dead_tup with stale/NULL last_autovacuum confirms bloat as
established; without sustained stats, bloat stays directional.
Findings
Emit findings per schema/finding.schema.json. Examples:
M18.db.txid_wraparound_imminent—xid_ageapproaching the ceiling (severity:5,fail, axisperformance, confidenceestablishedTier-1 only /needs_apiTier-0,fixable: advisory— aggressiveVACUUM (FREEZE)).M18.events.autovacuum_disabled— autovacuum turned off on a churny table (severity:3,warn,establishedfrom DDL,fixable: proposed).M18.events.dead_tuple_bloat— high dead-tuple ratio (severity:3,warn,establishedTier-2 /needs_api,fixable: advisory).
Each finding: evidence.observed quotes the config/DDL line or the catalog/stats row verbatim;
verification.reproduce is the query above referencing $DATABASE_URL; expected_impact is banded +
confidence-tagged (no naked %).
Honesty
- Wraparound sev-5 requires live
age()evidence — never declare it imminent from files; absent a connection, emitneeds_api, which never caps. - Some "bloat" is normal steady-state headroom; flag it only against a churn pattern or measured dead
ratio, and mark file-only verdicts
directional. - Never quote a dead-tuple percentage or a days-to-wraparound number you cannot observe; magnitude is
banded only. Routine maintenance fixes (
VACUUM, autovacuum tuning) areproposed/advisory.
Remedy — keep updates HOT
On an update-heavy table, lowering FILLFACTOR (e.g. the default 100/90 → 70) leaves free space in
each page so an UPDATE can place the new row version on the same page as a HOT (Heap-Only Tuple)
update — which skips writing new index entries, curbing index write-amplification and slowing bloat
growth. HOT only fires when no indexed column changed and the page has room, so a lower FILLFACTOR plus
not over-indexing the churned columns is the lever. This is the positive counterpart to the autovacuum/bloat
findings above; emit it as fixable: proposed (a FILLFACTOR change rewrites only on subsequent updates).
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/hainrixz/claude-db/db-storage-bloat">View db-storage-bloat on skillZs</a>