sentry-checkin: run log for skilld-dev/skilld.dev

Every run of the sentry-checkin routine reports here, including the runs that found nothing and the runs that were skipped.

Close a proposal's own issue to reject it. Closing this one stops the log, not the routine.

The Harlan Agent Kit opened this issue automatically. It is not Harlan's own report.

2026-08-26T20:20:00.000Z — Skipped. This run was due more than 6 hours ago, so it was skipped.

The scheduled sentry-checkin runtime has no Sentry issue-read token, so every run reports missing_observability and cannot see open issues such as SKILLD-11.

Target: /home/harlan/.sentryclirc

Verify with: python3 /home/harlan/.local/share/harlan-github-agent/service/harlan-agent-kit/skills/sentry-checkin/scripts/sentry_api.py --org harlan-zw snapshot --project skilld --output /tmp/opencode/skilld.snapshot.json

Estimated to change 1 file.

skills-raw caches a missing marker and returns 404 when the upstream tree fetch fails, while skill-asset returns 503 without caching for the same failure.

Target: layers/registry/server/api/skills-raw/[...slug].get.ts

Verify with: pnpm vitest run test/unit/skills-raw-tree-outage.test.ts

Estimated to change 2 files.

When ungh.cc returns 404 for the repo tree (no cached index for the branch), skills-raw throws 503 Skill source is unavailable upstream and pages Sentry (SKILLD-11, event 2026-08-29T18:31Z on deployed release 7457344), even though skills.rendered_skill_path already records the SKILL.md path needed to fetch the raw file directly from raw.githubusercontent.com without any tree listing.

Target: layers/registry/server/api/skills-raw/[...slug].get.ts

Verify with: pnpm vitest run test/unit/skills-raw-tree-outage.test.ts

Estimated to change 2 files.

The skill-asset handler shares the skills-raw ungh.cc tree dependency and answers the same 503 Skill source is unavailable upstream when ungh has no tree for the branch, though skills.rendered_skill_path joined alongside s.assets already names the skill directory needed to build the raw URL directly.

Target: layers/registry/server/api/skill-asset/[...slug].get.ts

Verify with: pnpm vitest run test/unit/skill-asset-tree-outage.test.ts

Estimated to change 2 files.

Nitro's built-in route cache (defineCachedEventHandler) forwards its already-caught best-effort KV write failures to Sentry as unhandled errors (SKILLD-17, KV PUT failed: 429), while project policy in shared/server/cache.ts records the same failure class as a wide event, so the server Sentry config needs a beforeSend filter for this exact signature with the wide event kept as the visibility path.

Target: shared/sentry.ts + server/plugins/sentry.ts

Verify with: pnpm test:run test/unit/sentry-cache-write-filter.test.ts

Estimated to change 2 files.

retryD1MaintenanceRead only retries long-running import/export errors, so transient 'D1 DB is overloaded' read failures 500 across 24 Sentry issues (SKILLD-13/14/18/19/1A-1K/1M-1T) during crawler sweeps instead of being replayed with the existing jittered backoff.

Target: server/utils/db.ts (retry classifier alongside isLongRunningD1Maintenance)

Verify with: pnpm vitest run test/unit/platform-d1.test.ts

Estimated to change 2 files.

nuxt-skew-protection 1.5.2 closes its BroadcastChannel on app:error while the app:manifest:update hook can still fire, so a later postMessage throws InvalidStateError on /gh/ file pages (SKILLD-12); multi-tab broadcast is redundant here because updateStrategy polling already reloads each tab.

Target: nuxt.config.ts skewProtection block (set multiTab: false)

Verify with: pnpm build && rg --no-ignore -c "multiTab" .output/server | head -3

Estimated to change 1 file.

nitro cached-handler SWR writes to the KV_CACHE mount exceed the 1 write/sec per-key limit and captureError raises each rejected best-effort write as an error even though the request succeeds, so SKILLD-17 and SKILLD-1V are recurring noise rather than a live defect.

Target: shared/sentry.ts plus server/plugins/sentry.ts (add a beforeSend that drops cache-tagged 'KV PUT failed: 429' rejections)

Verify with: pnpm vitest run test/unit/sentry-observability.test.ts

Estimated to change 3 files.

Sentry check-in — skilld.dev (project `skilld`) 2026-09-04, read-only

Verdict: AMBER. 34 open issues. No new defect needs new code beyond one small fix. Delivery is the real problem: production last deployed 2026-09-02T01:31Z and sits 4 commits behind main, so the merged #119 upstream-retry fix is undeployed and SKILLD-1E still fires.

Scope and mapping

Inventory: not found. Checked $SITES_FILE (unset), ~/SITES.md, ~/sites/SITES.md. Ran read-only against the user-named site. No edits, commits, pushes, or Sentry mutations.
The request named org skilld-dev; no such org exists. Sole org harlan-zw used. Project skilld (4511781692506112) maps to the skilld-dev/skilld.dev checkout. No unmatched projects, no zero-issue sites in scope.
Snapshot: ~/.local/state/sentry-checkin/run-jvGq09zU/skilld.snapshot.json. 34 unresolved ids, checksum c70fe319c58b…, duplicate_ids_dropped empty. Bundles reused from run run-BKLXtjvF (identical snapshot checksum), so evidence is from the same frozen set.
History: 34 new, 0 recurring, 0 unclosed. First recorded run for this project. History file: ~/.local/state/sentry-checkin/history.tsv. Nothing recorded this turn (read-only).

Deploy-state finding

wrangler deployments list: newest production deployment 2026-09-02T01:31:45Z, one version at 100%. The newest SKILLD-1E event (2026-09-03T17:50Z) reports release 1f5dcb9, the build that deployment carried. Production serves 1f5dcb9.
main is 4 commits ahead: eafbfd5 (fix(registry): retry a transient ungh tree read before answering 503 #119 tree-retry), 827faaf (fix(copy): align site copy with v3 run-first positioning #112), 8fdb2b2 (chore(digest): pause the AI summary behind an explicit flag #121), 4fcf0d9 (chore(routines): schedule the daily check-in #125).
Sentry holds a release 827faaf (created 2026-09-02T02:35Z, 0 events) that no Cloudflare deployment carries. The deploy workflow builds (which creates the Sentry release) then deploys, smokes, and rolls back. Reading: the 827faaf deploy never landed or was rolled back. GitHub Actions state unreadable this run (401).
Consequence: SKILLD-1E stays blocked on deploy. Do not resolve it yet.

Triage, one row per frozen id

D1 overload burst — 20 ids, fixed-via-candidate + covered. SKILLD-13,14,18,19,1A,1B,1C,1D,1F,1G,1H,1J,1K,1M,1N,1P,1Q,1R,1S,1T, all D1_ERROR: D1 DB is overloaded, on read endpoints, in 4-5 second crawler-sweep windows (Aug 31 06:06Z, Sep 2 09:17Z). The known ledger fingerprint, exit condition still unmet. No layer retries this message: @harlan-zw/nuxt-cloudflare TRANSIENT_D1_SIGNALS lacks it, and server/utils/db.ts retries only long-running import/export. One PR adds it to the replayable-read retry in server/utils/db.ts. The owning fix covers all 20 ids if a later sweep confirms.
KV PUT 429 — SKILLD-17, SKILLD-1V: expected. Nitro route-cache writes are best-effort; the setItem rejection is caught and reported via captureError (single Object.setItem frame, auto.function.nuxt.nitro). The request succeeds, the write is dropped. KV allows one write per key per second. No code change.
Skill source unavailable — SKILLD-11 (256 events, last 2026-08-31T22:10Z): already-fixed. Fixed by 24a14b5 (fix(registry): answer 410 for gone sources on skill-files and skill-asset #109), carried by the deployed 1f5dcb9 build. Last event precedes that deployment; quiet since. Resolve --in-release 1f5dcb92378e….
Skill source unavailable — SKILLD-1E (16 events, last 2026-09-03T17:50Z): blocked. Fix merged as eafbfd5 (fix(registry): retry a transient ungh tree read before answering 503 #119); production does not carry it. Next action: land the pending deploy of main, confirm one clean window, then resolve in that release.
Long-running export — SKILLD-15, SKILLD-16: third-party. Cloudflare D1 maintenance window. The e3045fc retry (4 attempts, ≤~480ms backoff) is deployed but cannot clear a minutes-long maintenance inside a request. Counts 2 and 1, stale since 2026-08-20.
Long-running import — SKILLD-10: already-fixed. Fixed by e3045fc (fix(d1): recover transient read failures #26); last event 2026-08-13T14:08Z predates it.
KV GET 500 — SKILLD-S: already-fixed. Fixed by 04dc7c7 (fix(registry): stop a KV outage 500ing every /gh page #40, readCache); last event 2026-08-18, quiet 16 days.
OG curator crash on all-digit login — SKILLD-V, SKILLD-W: already-fixed. og-props coercion landed in 56b6c56; ledger confirmed 2026-08-18; quiet since 2026-08-12.
Too many SQL variables — SKILLD-X: already-fixed. ba8ede4 (fix(registry): keep D1 reads within platform limits #21) + 0349755 (fix(registry): guard the cluster page against the D1 parameter cap #50) chunking; quiet since 2026-08-13.
D1_RESET_DO — SKILLD-Y: already-fixed. withD1ResetRecovery + e3045fc; one event, 2026-08-12.
withSession is not a function — SKILLD-Z: already-fixed. Nuxt Content route removed in 97a0d39; quiet since 2026-08-13.
BroadcastChannel closed — SKILLD-12: third-party. nuxt-skew-protection multi-tab plugin, 1 event, 2026-08-19. Upstream module owns the channel lifecycle; no local fix.

Proposed actions

Land the deploy of main and find why 827faaf never served. This unblocks SKILLD-1E and the SKILLD-11 resolution.
One PR: add D1 DB is overloaded to the replayable-read retry matcher in server/utils/db.ts, with a failing-first unit test in test/unit/platform-d1.test.ts. See the Candidate.
Next run (with write access): resolve SKILLD-11 in 1f5dcb9, resolve the six stale already-fixed ids (S, V, W, X, Y, Z, 10) per the release evidence above, and leave SKILLD-1E, the overload family, the expected and third-party rows open.

No Sentry issue was resolved this turn. No branch, worktree, or PR was created.

Confidence: 88/100. Verified: snapshot checksum, all 34 bundles, Cloudflare deployment history, release-to-event mapping, retry code paths. Unverified: GitHub CI state (401) and whether the 827faaf deploy failed smoke or never ran.

Sentry check-in — skilld.dev (read-only)

Verdict: AMBER. 34 open issues on project skilld (org harlan-zw), all new to this run's ledger. 20 of 34 are one live D1-overload family with events through 2026-09-04; one small code fix covers it. Most remaining ids are already fixed on main and are stale. Production lags HEAD, which holds one merged fix this family needs.

Scope

Site: skilld.dev ↔ Sentry project skilld, checkout /home/harlan/sites/skilld.dev.harlan-agent-routine-sentry-checkin-2026-09-04-7d58f4eaff46 (default branch, read-only this turn).
Site inventory: $SITES_FILE unset, ~/SITES.md missing, ~/sites/SITES.md missing — the repo's own docs/ops/triage-ledger.md served as prior-disposition evidence.
Snapshot: 34 issues, checksum c70fe319c58bd34493a8af5d377d0e3da26353b8af7db89ee0f8d788c0a32be5, no dropped duplicates, all states new. Run dir: ~/.local/state/sentry-checkin/run-sPenhpR7.
Production release 1f5dcb92 (deployed 2026-09-02T01:31Z). HEAD 5d2e6519 is two days ahead; PR fix(registry): retry a transient ungh tree read before answering 503 #119 (ungh tree-read retry) is on HEAD and not in production.

Findings by cluster

Cluster	Ids	Verdict
D1 overload bursts (4s crawler sweeps)	SKILLD-13,14,18,19,1A,1B,1C,1D,1F,1G,1H,1J,1K,1M,1N,1P,1Q,1R,1S,1T (20)	Live defect. Neither `retryD1MaintenanceRead` in `server/utils/db.ts` nor `@harlan-zw/nuxt-cloudflare`'s `TRANSIENT_D1_SIGNALS` lists the overload message, so replayable reads fail instead of retrying. Fix: extend the repo retry to replayable reads on this error. Candidate below.
KV PUT 429 in Nitro route cache	SKILLD-17, 1V	Expected degradation. The error is Nitro's internal `cachedEventHandler` KV write (`[cache] Cache write error`), caught and logged; the request succeeds. Repo `writeCache`/`readCache` already degrade best-effort on their paths. No code change ≤5 files without muting, which is forbidden. Disposition `expected`; wide events carry the signal.
Skill source unavailable upstream	SKILLD-11 (256), 1E (16)	Blocked on deploy. #109/#119 fixed the gone-source 410s and added the tree retry; #119 merged 2026-09-02 12:23 AEST, one hour after the production release was cut. SKILLD-1E's post-deploy events (last 2026-09-03T17:50Z) land on the un-deployed fix.
KV GET 500	SKILLD-S	Already fixed: `readCache` landed 2026-08-19 (#40), in production. Last event 2026-08-18T06:24Z precedes it. Resolve in the carrying release.
Long-running D1 import/export	SKILLD-10, 15, 16	Already fixed: read retry landed 2026-08-14/#26 and 2026-08-21/#55, in production. Last events precede the deploys.
Too many SQL variables on /api/clusters	SKILLD-X	Already fixed: parameter-cap guard 2026-08-20 (#50), in production. Last event 2026-08-13.
D1_RESET_DO	SKILLD-Y	Already fixed: `withD1ResetRecovery` landed 2026-08-14 (#26). Single event 2026-08-12.
`withSession is not a function` on content query	SKILLD-Z	Already fixed: withSession support in the maintenance-tolerant D1 wrapper landed 2026-08-14 (#26), one day after the last event. No recurrence in 3+ weeks.
Og-image numeric handle (`e58.split`)	SKILLD-V, W	Already fixed 2026-08-18 (`56b6c56`, og-props coercion), in production; last event 2026-08-12.
BroadcastChannel closed	SKILLD-12	Already fixed: nuxt-skew-protection 1.5.2 upgrade landed 2026-09-01 (`b6c4caf`), in production, with guard test. Last event 2026-08-19.

Proposed actions

One PR: overload-aware retry for replayable D1 reads (candidate). Expected to cut the 20-id overload family to noise and let the next run resolve them as fixed.
Deploy HEAD so fix(registry): retry a transient ungh tree read before answering 503 #119 ships; then resolve SKILLD-11/1E on evidence per the closed rows' pattern.
Next mutating run: resolve the stale already-fixed ids (S, 10, 12, 15, 16, V, W, X, Y, Z) in the releases that carried their fixes.
No ledger recorded, no resolves applied, no PR opened this turn: read-only.

2026-09-05T20:20:00.000Z — Skipped. This run was due more than 6 hours ago, so it was skipped.

Sentry check-in: skilld-dev/skilld.dev (read-only)

Verdict: 2 live defect classes need code, 11 already-fixed ids wait on Sentry resolution once deploy state is confirmed, 0 regressions. Frozen snapshot skilld.snapshot.json (checksum c70fe319c58bd34493a8af5d377d0e3da26353b8af7db89ee0f8d788c0a32be5, no duplicate ids dropped), 34 open issues, all new (no prior run history). No mutations made: no branch, no PR, no ledger recorded; history at /home/harlan/.local/state/sentry-checkin/history.tsv stays untouched.

Coverage (34/34)

Live, needs a PR (22 ids).

20 ids, D1_ERROR: D1 DB is overloaded. Requests queued for too long. (7683331480, 7683331513, 7701851775, 7701851777, 7701851783, 7701851784, 7701851793, 7701851846, 7706489539, 7706489557, 7706489559, 7706489573, 7706489580, 7706489596, 7706489601, 7706489615, 7706489624, 7706489639, 7706489647, 7706489648; ~347 events). Latest events ran release 1f5dc / 11fe9b8 (2026-08-31, 2026-09-02), which already carries every shipped mitigation (skill-related caching, canonical links, indexes). Verified missing: neither retryD1MaintenanceRead in server/utils/db.ts nor TRANSIENT_D1_SIGNALS in @harlan-zw/nuxt-cloudflare/d1 matches the overload message, so a replayable read that hits a full D1 queue fails on first attempt. Candidate 1 fixes the category and covers all 20 ids.
2 ids, KV PUT failed: 429 Too Many Requests (7701833213, 7706553864). Both postdate the writeCache fix (3ab8ea5, 2026-08-03). The 2026-09-02 bundle carries the breadcrumb [cache] Cache write error. Error: KV PUT failed: 429 Too Many Requests, which is Nitro 2.13.4's route cache (defineCachedEventHandler) writing hot keys straight to KV_CACHE; nitro catches the rejection but captureError raises a fresh Sentry issue per hot path on every crawler sweep. Candidate 2 fixes the category at the storage mount.

Already fixed at HEAD, resolution evidence only (11 ids). Every one of these last fired on a release that predates the fix commit (release names verified with git cat-file -e):

7662446055 KV GET 500 → fixed by readCache (04dc7c7, fix(registry): stop a KV outage 500ing every /gh page #40); last event 2026-08-18 on 2e4f9d6.
7666848469 + 7666848478 OG image e58.split → fixed by og-props.ts (56b6c56); last events 2026-08-12 on d2616c5.
7667355409 too many SQL variables on /api/clusters → fixed by the parameter-cap guard (0349755, fix(registry): guard the cluster page against the D1 parameter cap #50); last event 2026-08-13 on df28e5e.
7667379976 D1_RESET_DO → fixed by withD1ResetRecovery (e3045fc, fix(d1): recover transient read failures #26); last event 2026-08-12 on e0fe519.
7669649767 e57.withSession is not a function → 2 events, silent since 2026-08-13 on b37cf91, one day before fix(d1): recover transient read failures #26 rewired the D1 proxy; watch for one recurrence before further work.
7669999882, 7683331537, 7683331572 long-running import/export → fixed by the read retry (8ea5861, fix(d1): retry reads during exports #55, 2026-08-21); last events 2026-08-20 on aa64972, hours before the fix.
7676028614 Skill source is unavailable upstream (256 events) → fixed by the gone-source 410 boundary (fix(registry): answer 410 for gone sources on skill-files and skill-asset #109, 24a14b5c); last event 2026-08-31T22:10Z on d4e951e.
7705072099 same message (17 events through 2026-09-05T01:14Z) → the whole issue ran on release 1f5dc, built 2026-09-02T01:23Z, one hour before fix(registry): retry a transient ungh tree read before answering 503 #119 (eafbfd5, ungh tree retry) merged; production evidently sat on the pre-fix(registry): retry a transient ungh tree read before answering 503 #119 build while deploy checks were unstable (fix(ci): stabilize main deploy checks #132, 2026-09-04). already-fixed once the post-fix(registry): retry a transient ungh tree read before answering 503 #119 deploy is confirmed serving; blocked on that deploy confirmation if it has not happened.
7680671269 BroadcastChannel postMessage → fixed upstream in nuxt-skew-protection 1.5.2, installed and guarded by test/unit/skew-protection-multi-tab.test.ts.

Proposed actions

Candidate 1: extend the replayable-read retry to the D1 overload class (server/utils/db.ts, 2 files). Covers the 20 overload ids.
Candidate 2: mount a failure-tolerant driver on the cache storage so nitro route-cache KV rejections degrade to a miss with a wide event instead of a captured error (nuxt.config.ts + driver + test, 3 files). Covers the 2 KV PUT ids.
Next run: resolve the 11 already-fixed ids per the skill's release rules once the current HEAD deploy is confirmed, and re-check 7705072099 for post-fix(registry): retry a transient ungh tree read before answering 503 #119 growth.

The maintenance read retry in retryD1MaintenanceRead only matches long-running import and export messages, so a D1 overload rejection fails a replayable read on its first attempt and 20 live endpoint issues keep growing.

Target: server/utils/db.ts

Verify with: pnpm vitest run test/unit/platform-d1.test.ts

Estimated to change 2 files.

Nitro's route cache writes hot keys straight to the KV_CACHE binding, so each 429 from the one-write-per-second limit is captured to Sentry as a fresh error issue per hot path during a crawler sweep.

Target: nuxt.config.ts

Verify with: pnpm vitest run test/unit/nitro-cache-storage.test.ts

Estimated to change 3 files.

Sentry check-in: skilld.dev (project `skilld`)

Verdict: backlog is 34 issues, all new to the ledger, zero prior dispositions. Every issue is accounted for. Two live noise sources get one small filter PR each. Everything else is already fixed in code; most of those fixes deployed on 2026-09-07 and just need a Sentry resolve with release evidence on the next mutating run.

Scope and method

Org: harlan-zw (the only organization; skilld-dev is the GitHub owner, not a Sentry org). Project skilld, matched to skilld.dev through sentry.client.config.ts and shared/sentry.ts (DSN project id 4511781692506112).
Site inventory: no inventory exists. Paths checked: $SITES_FILE (empty), ~/SITES.md, ~/sites/SITES.md. Scope taken from the request instead.
Snapshot: 34 issues, no duplicate IDs dropped, checksum c70fe319c58b.... Run dir: ~/.local/state/sentry-checkin/run-9gwt5G2X.
History: ~/.local/state/sentry-checkin/history.tsv is empty. Counts: 34 new, 0 recurring, 0 unclosed.
This run was read only. No worktree, no edits, no PR, no resolutions, no ledger record.

Triage, all 34 issues

Cluster	Issues (short IDs)	Disposition	Evidence
D1 overload	20: SKILLD-13, 14, 18, 19, 1A, 1B, 1C, 1D, 1F, 1G, 1H, 1J, 1K, 1M, 1N, 1P, 1Q, 1R, 1S, 1T	already-fixed	PR #158 (`f82a70a`, 2026-09-07) added `d1 db is overloaded` to `RETRYABLE_D1_READ_MESSAGES` in `server/utils/db.ts`. All events are 2026-09-02 to 09-04, from releases `11fe9b8a` and `1f5dcb92`, which predate the fix. `f82a70a` is an ancestor of release `8b1af9f`, built 2026-09-07; a Cloudflare deployment followed at 06:14Z. Resolve in `8b1af9f` once its deploy SHA is confirmed.
KV PUT 429	2: SKILLD-17, 1V	already-fixed	PR #159 (`4025044`) routes Nitro cache writes through the tolerant KV mount and drops the forwarded error in `server/plugins/sentry.ts` `beforeSend`, via `isBestEffortCacheWriteError`. Events 2026-08-31 and 09-02 predate it. Same release evidence as above.
Long-running import/export	3: SKILLD-10, 15, 16	already-fixed	`8ea5861` (2026-08-21) retries reads during D1 maintenance; ancestor of release `11fe9b8a` (2026-08-31). Events 2026-08-13 and 08-20 come from `b37cf91e` and `aa64972d`.
Upstream unavailable 503	2: SKILLD-11, 1E	expected, filter proposed	`layers/registry/server/api/skill-files/[...slug].get.ts` throws 503 with Retry-After when `fetchUpstreamTree` reports `unavailable`. That is intended. The handler already emits an operational wide event for the outage, so the Sentry capture is duplicate noise. Last event 2026-09-05, so this recurs. Candidate 1.
SQL variables	1: SKILLD-X	already-fixed	`ba8ede4` (2026-08-13 18:27 AEST) chunked the `/api/clusters` selectors at `D1_BOUND_PARAMETER_LIMIT`. The one event fired 08:35Z the same day from `df28e5ec`, minutes before the fix. No recurrence.
KV GET 500	1: SKILLD-S	already-fixed	`04dc7c7` (PR #40, 2026-08-19) made `readCache` treat a failed read as a miss. The single event, 2026-08-18 16:24 AEST, ran on `2e4f9d6b`, built 15:16 AEST the day before the fix.
D1_RESET_DO	1: SKILLD-Y	already-fixed	`e3045fc` (2026-08-14) wraps reads in `withD1ResetRecovery`. Event 2026-08-12 from `e0fe519a`.
withSession TypeError	1: SKILLD-Z	already-fixed	Same commit `e3045fc` added the `withSession` proxy in `createMaintenanceTolerantDatabase`. Event 2026-08-13 from `b37cf91e`.
Curator OG image crash	2: SKILLD-V, W	already-fixed	Numeric-login coercion landed on `56b6c56` (2026-08-12); repo triage ledger row confirms exit condition met 2026-08-18. Events 2026-08-12 from `d2616c5e`.
BroadcastChannel race	1: SKILLD-12	third-party, mitigation proposed	The error comes from `nuxt-skew-protection`'s multi-tab client plugin, not repo code. One event, 2026-08-19. A client `beforeSend` filter is the safe local mitigation. Candidate 2.

Proposed actions

Candidate 1: drop expected upstream-unavailable 503s in the server beforeSend filter. Files: shared/sentry.ts, server/plugins/sentry.ts, one test file. Covers SKILLD-11, SKILLD-1E.
Candidate 2: drop the benign closed-channel postMessage error in a client beforeSend filter. Files: shared/sentry.ts, sentry.client.config.ts, one test file. Covers SKILLD-12.
Next mutating run: confirm release 8b1af9f is the deployed SHA, then resolve the 29 already-fixed issues listed above in their carrying releases. Do not resolve expected or third-party rows.

Unmatched projects: none. Unmatched inventory sites: none in scope. Zero-issue sites: not applicable, single project.

Confidence: 88/100. Verified: snapshot checksum, per-issue bundles, fix commits and their release ancestry, and the Cloudflare deployment timestamp. Unverified: the exact SHA inside the 2026-09-07 06:14Z deployment, and that the Sep 7 releases receive no new events under live traffic.

The skill-files, skill-asset and skills-raw handlers throw their intended upstream-outage 503 as an unhandled error, so Sentry records SKILLD-11 and SKILLD-1E although fetchUpstreamTree already reports the outage as an operational wide event.

Target: shared/sentry.ts

Verify with: pnpm vitest run test/unit/sentry-cache-write-filter.test.ts

Estimated to change 3 files.

The client Sentry config reports a benign tab-close race in nuxt-skew-protection's multi-tab plugin (SKILLD-12, InvalidStateError: Channel is closed) as an error.

Target: shared/sentry.ts

Verify with: pnpm vitest run test/unit/sentry-cache-write-filter.test.ts

Estimated to change 3 files.

Sentry check-in, skilld.dev, 2026-09-08 (read-only)

Verdict: AMBER. 37 open issues, all frozen as new for this routine's history file. 14 are already-fixed in Sentry's eyes and only need resolving against deployed releases; 25 belong to two live families: D1 saturation during crawler sweeps (23 ids, events today) and upstream-blip 503s on skills-raw (SKILLD-1E, grew after the #119 retry fix). Five small cache candidates proposed below; no code was changed, committed, or pushed.

Scope and method

Organization harlan-zw, project skilld, mapped to this checkout via nuxt.config.ts sentry config. Sole org, no ambiguity.
No site inventory exists at $SITES_FILE, ~/SITES.md, or ~/sites/SITES.md; scope taken from the request plus the repo's tracked sentry config.
Snapshot: 37 unresolved issues, sha256 97434734639e5664a580131ae3c0d34891423505959af8d88ab98b3de16514a0, no duplicate ids dropped. One event bundle per issue fetched. History file ~/.local/state/sentry-checkin/history.tsv is empty: first run of this routine, so new 37, recurring 0, unclosed 0. The repo's own docs/ops/triage-ledger.md was used as prior-art evidence.
Deployed release at snapshot time is 103991d, equal to origin/main HEAD.

Dispositions

Already-fixed, resolve next run in the carrying release after confirming the deploy (14 ids, all with last-seen before their fix deployed and 5+ days clean): SKILLD-S (readCache, shared/server/cache.ts), SKILLD-V and SKILLD-W (og-props numeric-handle fix, 56b6c56), SKILLD-Z (content route removed, 97a0d39), SKILLD-X (D1 SQL variable chunking, ba8ede4), SKILLD-Y (D1 reset recovery, #26), SKILLD-10 (import retry, #55), SKILLD-15 and SKILLD-16 (export retry, #55), SKILLD-12 (BroadcastChannel skew fix #66), SKILLD-11 (gone-source 410 on skill-files/skill-asset, #109).

Watch, then resolve: SKILLD-17 and SKILLD-1V (KV PUT 429). The beforeSend drop (#176) and route-cache wide-event resolution (#159) deployed 2026-09-07/08, after both ids' last events. Expect the next burst window to be silent; then resolve.

Live, D1 overload family (23 ids): SKILLD-13, 14, 18, 19, 1A, 1B, 1C, 1D, 1F, 1G, 1H, 1J, 1K, 1M, 1N, 1P, 1Q, 1R, 1S, 1T, 1W, 1X, 1Y. Still firing today on the current release despite #158 (D1 read retry) and #168 (stale fallback on skill-related). Mechanism per the ledger: crawler sweeps of unique cold keys; no cache helps a never-cached key. Root mitigation is capacity or crawler shaping, which is a decision, not a small fix; the five candidates below remove the remaining uncached and short-TTL hot paths and would have covered 9 of today's 23 culprit endpoints.

Live, skills-raw outage 503s: SKILLD-1E. Five events 2026-09-03 to 09-05 across four repositories, after #119's tree retry deployed. Matches the ledger's own over-firing exit condition; candidate 5 is the mitigation.

Blocked: none.

Proposed actions

Five candidates, each one small PR: cache the org profile endpoint, cache the user skills endpoint, add stale serving to the tag profile cache, add stale serving to both cluster endpoints, and serve the stale skills-raw body when upstream blips. Resolve the 14 already-fixed ids in their carrying releases on the next non-read-only run. No Sentry resolutions were applied this run; nothing was muted.

Confidence: 91/100. Verified live Sentry state, deployed release, fix lineage in git, and handler cache shapes by reading code. Untested path: whether ungh (not skilld) was down during the 09-03 to 09-05 SKILLD-1E events; the events are consistent with either an ungh outage or classifier over-firing.

GET /api/orgs/:owner is a bare defineEventHandler with no route or read-through cache, so every crawler hit runs five-plus D1 queries plus per-repo upstream fetches, which is the recurring D1 overload culprit (SKILLD-1M, SKILLD-1N, and the 2026-08-05 orgs family).

Target: layers/registry/server/api/orgs/[owner].get.ts

Verify with: pnpm vitest run test/unit/org-profile-cache.test.ts

Estimated to change 2 files.

GET /api/users/:login/skills is a bare defineEventHandler with no cache, so each crawler hit repeats the same D1 reads and landed as SKILLD-1K in the D1 overload burst.

Target: server/api/users/[login]/skills.get.ts

Verify with: pnpm vitest run test/unit/user-skills-cache.test.ts

Estimated to change 2 files.

The tag profile handler caches for 60 seconds with swr disabled on a heavy FTS query plus per-row subqueries, so crawler sweeps recompute it continuously (SKILLD-19, SKILLD-1C, SKILLD-1H).

Target: layers/registry/server/api/tags/[slug].get.ts

Verify with: pnpm vitest run test/unit/tag-profile-cache.test.ts

Estimated to change 2 files.

Both cluster handlers cache for 60 seconds with swr disabled, so crawler sweeps re-run the cluster index D1 queries every minute (SKILLD-1A, recurring burst member).

Target: layers/registry/server/api/clusters/index.get.ts and layers/registry/server/api/clusters/[slug].get.ts

Verify with: pnpm vitest run test/unit/cluster-cache.test.ts

Estimated to change 3 files.

The raw endpoint keeps a fetched body for only 5 minutes with no stale fallback, so an upstream blip after expiry answers 503 retry-after on the npx skilld run surface although a last-good copy existed minutes earlier (SKILLD-1E kept growing after #119).

Target: layers/registry/server/api/skills-raw/[...slug].get.ts

Verify with: pnpm vitest run test/unit/skills-raw-tree-outage.test.ts

Estimated to change 2 files.

Sentry check-in, skilld-dev/skilld.dev, 2026-09-09T20:20:00.000Z

GREEN. 42 frozen issues, all covered in the audited ledger. 9 resolved in the verified deployed release 103991d0b34830f98e5b361c4c8665d8f0d75aaa. 33 stay open as third-party Cloudflare platform failures with mitigations already deployed. No code repairs needed, so no Candidates and no PR. Repository unchanged.

Scope and access

Repository: skilld-dev/skilld.dev, controller worktree skilld.dev.harlan-agent-routine-sentry-checkin-2026-09-09-9f3c9c18b5ab on 103991d (= origin/main).
Organization: harlan-zw. Project: skilld.
Project mapping: DSN project id 4511781692506112 in shared/sentry.ts:1 matches Sentry project skilld (id 4511781692506112). One project, exact match, no unmatched projects.
sentry-cli installed and authenticated. No script failures. No duplicate ids dropped.

Counts

Metric	Value
Frozen issues	42
Ledger rows	42 (100% coverage, audit pass)
new	42
recurring	0
unclosed	0
already-fixed (resolved this run)	9
third-party (left open)	33
Candidates	0

First routine run against this history file, so every id tags new. Prior manual triage in docs/ops/triage-ledger.md was used as leads only.

Deployment evidence

Release 103991d0b34830f98e5b361c4c8665d8f0d75aaa:

Released 2026-09-08T07:40:06Z (Sentry release record).
Wrangler deployment record 2026-09-08T07:40:37Z, 100% traffic.
Fresh production events landed on this release through 2026-09-09T02:11Z, so it was live.
Live probe 2026-09-09: https://skilld.dev/ 200, /api/clusters 200.
Every fix commit below verified as an ancestor of this release with git merge-base --is-ancestor.

Resolved this run (9), all `--in-release 103991d0b34830f98e5b361c4c8665d8f0d75aaa`

Issue	Fix evidence
SKILLD-S `7662446055` KV GET 500	`readCache` treats a rejected KV read as a miss (`acbd24c` #40). Last event 2026-08-18 precedes the fix.
SKILLD-V `7666848469` OG island 500, digit login	`og-props` coercion (`56b6c56`), test-guarded. Last event 2026-08-12T01:32Z.
SKILLD-W `7666848478` `e58.split is not a function`	Same root cause and fix as SKILLD-V.
SKILLD-X `7667355409` too many SQL variables	`findClusterIndexRows` chunks at the D1 100-parameter cap (`ba8ede4` #21, `0349755` #50). Last event 2026-08-13; live endpoint 200 today.
SKILLD-Z `7669649767` `withSession is not a function`	db proxy forwards `withSession` (`e3045fc` #26, 2026-08-14). Last event 2026-08-13T13:01Z.
SKILLD-11 `7676028614` skill source unavailable upstream	Gone sources answer 410 (`24a14b5c0` #109), tree reads retry (`eafbfd5` #119), residual capture dropped in `beforeSend` (`c318148` #176). Last event 2026-08-31T22:10Z.
SKILLD-17 `7701833213` KV PUT 429 captured	Route-cache KV write failures resolve as wide events (`4025044` #159); `beforeSend` drops the signature. Last event 2026-08-31T05:48Z.
SKILLD-1E `7705072099` skill source unavailable upstream	Same class as SKILLD-11: retry (`eafbfd5`) + capture filter (`c318148`). Last event 2026-09-05T01:14Z, before this release.
SKILLD-1V `7706553864` KV PUT 429 captured	Same class as SKILLD-17: wide events (`4025044`) + capture filter. Last event 2026-09-02T10:04Z.

Resolution plan ran before --apply. All 9 re-read through the API after applying: status=resolved, statusDetails.inRelease=103991d0b34830f98e5b361c4c8665d8f0d75aaa. Sentry reopens any of these as a regression if the error returns.

Left open (33), all `third-party`, never resolved

28 D1 overload ids (SKILLD-13/14/18/19/1A/1B/1C/1D/1F/1G/1H/1J/1K/1M/1N/1P/1Q/1R/1S/1T/1W/1X/1Y/1Z/20/21/22/23): Cloudflare D1 capacity rejection during crawler and ingest bursts. All frames terminate in cloudflare-internal d1-api. Mitigations deployed: route caching (d56a917, bb607cc) and the replayable-read retry (f82a70a fix(d1): retry replayable reads on D1 overload #158). Events at 2026-09-09T02:11Z continue after the retry fix, so the residual is platform capacity. Exit condition unchanged: a crawler sweep produces no burst.
SKILLD-10/15/16: D1 refuses reads while a long-running import or export holds the database. Platform lock, no local frame.
SKILLD-Y: Cloudflare D1 durable-object reset (D1_RESET_DO). Platform transient.
SKILLD-12: nuxt-skew-protection BroadcastChannel postMessage after close. Stack entirely in the dependency runtime, one event, 2026-08-19, no recurrence.

Candidates

None. No disposition this run needed a code repair. The repository is unchanged (0 tracked modifications), no branch, no PR.

Artifacts

Run directory: /home/harlan/.local/state/sentry-checkin/runs/766422606816d7c50ff67eb7914ba507b0e79c3e7ba9b3044ca945b81a8ff69b
Snapshot: skilld.snapshot.json (issue_ids_sha256 8b1beecf19fbb0bb2c0a867d4dbb71315d8177173b3407f81d5d1dea9b8dc644)
History report: skilld.history.json
Bundles + audit manifest: artifacts/skilld/manifest.json
Ledger: artifacts/ledger.tsv (42 rows, ledger_sha256 978278b8b59234846a86e3d7c72292704ce3d4e2ab477b0acc9a9312e0893a6a, audit pass)
Resolution plan / apply / verify: skilld.resolve-plan.json, skilld.resolve-apply.json, skilld.resolve-verify.txt
Run history: /home/harlan/.local/state/sentry-checkin/history.tsv (42 rows appended for run id skilld-dev/skilld.dev:sentry-checkin:2026-09-09T20:20:00.000Z)

Confidence: 95/100. Verified end to end: frozen snapshot, 42-row audited ledger, resolution plan, applied resolutions, and per-issue API status checks. The untested remainder is future behaviour: a recurrence reopens any resolved id, and the D1 overload family depends on Cloudflare capacity, not on this run.

Sentry Check-in: skilld-dev/skilld.dev — 2026-09-10T20:20:00.000Z

Verdict: Backlog is one Cloudflare D1 capacity class plus one stale upstream bug. This run resolved SKILLD-12 in the verified live release. The 33 D1 rows stay open as third-party: the deployed read-retry mitigation is live and cannot remove Cloudflare capacity errors. No code repairs this run.

Scope and mapping

Organization: harlan-zw (sole org). Project: skilld.
Mapping proof: tracked DSN in shared/sentry.ts:1 points at project id 4511781692506112, which is the skilld project in Sentry. Not name similarity.
Unmatched projects: none. Unmatched inventory sites: none.
duplicate_ids_dropped: none.
Worktree untouched: git status --porcelain clean. No commits, pushes, PRs, or deploys.

Coverage

Frozen issues: 34 (skilld.snapshot.json, query is:unresolved).
Ledger rows: 34/34, every frozen ID has one evidence-backed row.
Audit: passed. Snapshot checksum ee5b3eb5…eaf5a equals audit checksum.
History states: new 1 (SKILLD-24), recurring 33, unclosed 0.

Resolved this run

Issue	Disposition	Resolved in release	Proof
SKILLD-12 (7680671269) BroadcastChannel postMessage on closed channel	`already-fixed`	`103991d0b34830f98e5b361c4c8665d8f0d75aaa`	Upstream fix installed in `b6c4caf` (catalog → nuxt-skew-protection 1.5.2, local patch dropped), which is an ancestor of release `103991d0b`. That release is proven deployed: it carried production events dated 2026-09-10T15:36. Last SKILLD-12 event 2026-08-19 predates the fix; no recurrence across three weeks. Guarded by `test/unit/skew-protection-multi-tab.test.ts`.

Plan (resolve-plan.json) ran before --apply (resolve-applied.json); API re-read (resolve-verify.json) confirms status: resolved, inRelease: 103991d0b….

Unresolved, with reasons

33 issues → third-party (Cloudflare D1 capacity/maintenance):
- 29 × D1_ERROR: D1 DB is overloaded (SKILLD-13, -14, -18, -19, -1A, -1B, -1C, -1D, -1F, -1G, -1H, -1J, -1K, -1M, -1N, -1P, -1Q, -1R, -1S, -1T, -1W, -1X, -1Y, -1Z, -20, -21, -22, -23, -24).
- SKILLD-Y: D1_ERROR: {"D1_RESET_DO":true} (Cloudflare Durable Object reset).
- SKILLD-10: D1 long-running import lock. SKILLD-15, -16: D1 long-running export lock.
- Evidence: all three signatures are already retried with backoff by retryReplayableD1Read (server/utils/db.ts:133, deployed in f82a70a, present in live release 103991d0b), yet fresh failures on 2026-09-10 still surfaced from that release. The root cause is Cloudflare capacity, no local code can remove it. Mitigation decision: keep the read retries; accept residual 500s during D1 overload windows. Never resolved in Sentry; a regression would still surface.
No issue was left blocked. No issue was muted.

Code repair Candidates

None. Every open issue is third-party with its mitigation already deployed; the one fixable issue was already fixed upstream and is now resolved in Sentry.

Artifacts

Run dir: /home/harlan/.local/state/sentry-checkin/runs/c20bca4f77bd529927338dc052f0412a72b7984798eb146372a330ce01b86f56
Snapshot: skilld.snapshot.json
History report: skilld.history.json
Bundles/manifest: skilld.bundles.json/ (+ manifest.json)
Ledger: ledger.tsv — audit: audit.json
Resolution evidence: resolve-plan.json, resolve-applied.json, resolve-verify.json
History: /home/harlan/.local/state/sentry-checkin/history.tsv (34 rows appended for run skilld-dev/skilld.dev:sentry-checkin:2026-09-10T20:20:00.000Z)

Confidence: 93/100. Verified end to end: snapshot checksums, audit, resolve plan → apply → API re-read, history append. Untested: none within this run's scope; the D1 third-party verdict rests on Cloudflare-side behavior this run cannot exercise.

Sentry check-in: skilld-dev/skilld.dev, 2026-09-11

Verdict: backlog unchanged and all third-party. One new Nuxt error-page crash, no local fix. Zero eligible Sentry resolutions. No code Candidates, no PR.

Scope

Repository: skilld-dev/skilld.dev, controller worktree skilld.dev.harlan-agent-routine-sentry-checkin-2026-09-11-7efb6baa95de (read only).
Organization: harlan-zw (sole org). Project: skilld, matched by tracked DSN in shared/sentry.ts:1 (project ID 4511781692506112). No unmatched projects, no exclusions.
Sentry access verified with sentry-cli info.

Coverage

Frozen issues: 33 (snapshot checksum c3e117218cc846bf3693c5a9732eff2eddd2da49f0c6e851f69c56ca9f550859).
Ledger rows: 33, every row dispositioned. Audit passed; ledger checksum 30badebbe33d0d4397f9c6ab100c207412bf691803ec26312dda60ad0e466543.
Duplicate IDs dropped by snapshot: none.
Counts: new 1, recurring 32, unclosed 0.

Dispositions

33 third-party:
- 32 D1 errors (D1_ERROR: D1 DB is overloaded / long-running import/export). Every stack terminates in cloudflare-internal:d1-api (D1DatabaseSession._sendOrThrow) behind Nitro handlers. Events span releases up to 103991d0b348 with latest event 2026-09-10, so the platform failure is still live. The app already retries through runRecoverable. No local fix exists; mitigation is keep-open monitoring and Cloudflare escalation if volume grows.
- 1 new SKILLD-25 (7726701629) TypeError: t.toString is not a function, single client event 2026-09-11 at /__nuxt_error.md, release 103991d0b348. Mechanism auto.function.nuxt.vue-error; every frame is nuxt@4.5.2 / vue@3.5.42 dist code, innermost nuxt-error-page.vue Lg.setup. The repo ships no error.vue and no local code references __nuxt_error; the .md source plugin only answers /gh/*/*.md. No local fix; mitigation is monitor and add a custom error.vue or upgrade Nuxt if it recurs.

Sentry resolutions

Eligible resolutions: 0. No fixed, covered, or already-fixed rows exist, so no --in-release plan ran and nothing was applied. All 33 rows stay open by rule (third-party is never resolved).
Muted issues: none.

Code repairs

Candidates: 0. No issue has a local root cause this run.
No repairs were previously rejected.

Artifacts

Snapshot: /home/harlan/.local/state/sentry-checkin/runs/2dc6ce5f4883f28e191b84cbc6fb2f559c2d5d4dd37daef9fa11396401e7771b/skilld.snapshot.json
Bundles + audit manifest: /home/harlan/.local/state/sentry-checkin/runs/2dc6ce5f4883f28e191b84cbc6fb2f559c2d5d4dd37daef9fa11396401e7771b/skilld.bundles.json/
History report: /home/harlan/.local/state/sentry-checkin/runs/2dc6ce5f4883f28e191b84cbc6fb2f559c2d5d4dd37daef9fa11396401e7771b/skilld.history.json
Ledger: /home/harlan/.local/state/sentry-checkin/runs/2dc6ce5f4883f28e191b84cbc6fb2f559c2d5d4dd37daef9fa11396401e7771b/ledger.tsv
Audit output: /home/harlan/.local/state/sentry-checkin/runs/2dc6ce5f4883f28e191b84cbc6fb2f559c2d5d4dd37daef9fa11396401e7771b/audit.json
Run history: /home/harlan/.local/state/sentry-checkin/history.tsv (33 rows appended for run skilld-dev/skilld.dev:sentry-checkin:2026-09-11T20:20:00.000Z, date 2026-09-11)

Script failures and missing evidence

None. No step failed. No deployment evidence was needed because no row qualified for resolution.

Sentry Check-in: skilld.dev (skilld) — 2026-09-12

Verdict: no action needed. All 32 open issues are already-attributed upstream classes (29x Cloudflare D1 capacity, 2x D1 export lock, 1x Nuxt error-page internals); zero eligible resolutions, zero code repairs this run.

Run identity

Routine run ID: skilld-dev/skilld.dev:sentry-checkin:2026-09-12T20:20:00.000Z
Mode: propose (scheduled)
Repository: skilld-dev/skilld.dev, controller worktree at HEAD 103991d0 (clean, unchanged)
Sentry: org harlan-zw, project skilld (matched via DSN project id 4511781692506112 in shared/sentry.ts:1)

Coverage

Frozen snapshot: 32 unresolved issues, 0 duplicate IDs dropped
Ledger: 32/32 rows dispositioned, audit passed
Checksum: snapshot c7c9d6c8...53e2c6 == ledger audit c7c9d6c8...53e2c6
History tags: 0 new, 32 recurring, 0 unclosed

Dispositions (all 32 rows)

Group	IDs	Disposition	Evidence
D1 overload (`D1 DB is overloaded. Requests queued for too long.`), 29 groupings	7683331480, 7683331513, 7701851775, 7701851777, 7701851783, 7701851784, 7701851793, 7701851846, 7706489539, 7706489557, 7706489559, 7706489573, 7706489580, 7706489596, 7706489601, 7706489615, 7706489624, 7706489639, 7706489647, 7706489648, 7719001379, 7719001402, 7719243634, 7720451298, 7720451335, 7720658362, 7720658378, 7720658525, 7723027152	third-party	handled=yes, runtime=cloudflare; last events 2026-09-02..09-10, all on the current release `103991d0`/`1f5dcb92`; mitigation (SWR inflight dedup, `shared/server/cache.ts`) already deployed; kept open as upstream capacity signal
D1 export lock (`Currently processing a long-running export.`)	7683331537, 7683331572	third-party	last event 2026-08-20 on `aa64972d`; manual D1 export lock, no local fix; quiet 3+ weeks
Nuxt error page TypeError	7726701629	third-party	single event 2026-09-11 on `103991d0`; all 18 frames inside vue/nuxt 4.5.2 deps, throw site `nuxt-error-page.vue` setup, zero app frames; mitigation: monitor, Sentry reopens on regression

Resolutions

Eligible (already-fixed / covered): none. Every issue's latest event landed on the currently deployed release 103991d0 (live events 28h before the run), so no issue is provably fixed by a deployed release.
Plan: empty; nothing applied to Sentry. Per contract, third-party rows are never resolved.

Code repairs

Candidates: none. No issue traces to app code; all throw sites are Cloudflare infra or Nuxt/Vue internals.
Repairs beyond the file limit: none.

Notes

Prior run (2026-09-11) dispositions were third-party for all 32; this run re-verified each against fresh event evidence and the repo's deployed mitigation before agreeing.
7706489539 remains the highest-volume issue (1030 events, last 2026-09-10). Volume is upstream; watch Cloudflare D1 status rather than the app.

Artifacts

Snapshot: /home/harlan/.local/state/sentry-checkin/runs/41d2b346f055602424dba3681bc9eb7b0b66cd6b14e649ff46d3228a0e15b0c9/skilld.snapshot.json
History report: /home/harlan/.local/state/sentry-checkin/runs/41d2b346f055602424dba3681bc9eb7b0b66cd6b14e649ff46d3228a0e15b0c9/skilld.history.json
Evidence bundles: /home/harlan/.local/state/sentry-checkin/runs/41d2b346f055602424dba3681bc9eb7b0b66cd6b14e649ff46d3228a0e15b0c9/skilld.bundles.json/ (32 bundles + manifest.json)
Ledger (audited): /home/harlan/.local/state/sentry-checkin/runs/41d2b346f055602424dba3681bc9eb7b0b66cd6b14e649ff46d3228a0e15b0c9/ledger.tsv
This report: /home/harlan/.local/state/sentry-checkin/runs/41d2b346f055602424dba3681bc9eb7b0b66cd6b14e649ff46d3228a0e15b0c9/report.md
History (appended 32 rows): /home/harlan/.local/state/sentry-checkin/history.tsv

Sentry check-in: skilld.dev/skilld.dev (org harlan-zw, project skilld)

Verdict: healthy triage, no code defects. All 32 frozen issues are upstream or unattributable; 31 are Cloudflare D1 capacity/export-lock failures surfaced through cloudflare-internal:d1-api, 1 is a single unattributable client TypeError inside minified Nuxt error-page internals. Zero eligible Sentry resolutions this run; zero code repair Candidates. Nothing was muted.

Scope and mapping

Repository: skilld.dev/skilld.dev, controller worktree, read-only (git status clean at end).
Sentry access: sentry-cli authenticated, sole org harlan-zw.
Project mapping: tracked DSN in shared/sentry.ts:1 (project id 4511781692506112) maps exactly to project skilld. No unmatched projects. No inventory exclusions apply.
Query: is:unresolved via sentry-cli, 32 issues frozen, no duplicate IDs dropped.

Counts

Frozen issues: 32; ledger rows: 32 (100% coverage); audit checksum c7c9d6c8d554ca3c83d4e62ce6f08854513ae031757e0ac21b0a4fc26e53e2c6 matches the snapshot.
History: new 0, recurring 32, unclosed 0.
Dispositions: third-party 32.
Resolved this run: none (no fixed/already-fixed/covered rows exist; third-party is never resolved).

Issue inventory (all third-party)

D1_ERROR: D1 DB is overloaded. Requests queued for too long. (29 issues: SKILLD-13, 14, 18, 19, 1A, 1B, 1C, 1D, 1F, 1G, 1H, 1J, 1K, 1M, 1N, 1P, 1Q, 1R, 1S, 1T, 1W, 1X, 1Y, 1Z, 20, 21, 22, 23, 24): every event stack originates at cloudflare-internal:d1-api D1DatabaseSession._sendOrThrow; application frames are ordinary query helpers (findSkill, findRelatedSkills, getGeneratedBatch, findSkillCommitSource, listDuplicateCandidateRows, tag/org/repo handlers). Largest: SKILLD-1F, 1047 events, still firing 2026-09-13. Evidence: upstream Cloudflare D1 capacity saturation under request bursts. Mitigation: cached handler paths and wide events absorb the blast; issues stay open to watch for regression; no mute.
D1_ERROR: Currently processing a long-running export. (2 issues: SKILLD-15, 16): same D1 seam; a long-running export holds the database. Last seen 2026-08-20; not recurring since. Mitigation: no mute, watch.
TypeError: t.toString is not a function (SKILLD-25, 1 event, 2026-09-11, Chrome 144 zh-CN, culprit /__nuxt_error.md): all 18 frames are minified Vue 3.5.42 / Nuxt 4.5.2 internals; last frame nuxt-error-page.vue Lg.setup; zero application frames. No local fix attributable; client sourcemaps do not match events (release naming gap noted in shared/sentry.ts comments). Mitigation: no mute, monitor recurrence.

Resolutions

Plan: empty (resolution-plan.txt in the run dir). No resolve call, no --apply, no status re-reads required. No issues are eligible while every row is third-party.

Code repair Candidates

None. A D1 retry layer would touch the shared platform DB seam and every write/read path, exceeding the small-repair bar and risking retry amplification during saturation; it is an infrastructure design decision, not a defect fix, so the D1 rows stay third-party. No prior rejections on file.

Artifacts

Snapshot: /home/harlan/.local/state/sentry-checkin/runs/059a385f9a1dc4393e399d7d1c507d3763703ca1d583b76dd0ad5541c760e3f0/skilld.snapshot.json
Bundles/manifest: .../skilld.bundles.json/manifest.json (+ 32 per-issue bundles)
Ledger: .../ledger.tsv; audit passed (32 rows, checksum match)
History: /home/harlan/.local/state/sentry-checkin/history.tsv (32 rows appended for run skilld-dev/skilld.dev:sentry-checkin:2026-09-13T20:20:00.000Z, 2026-09-13)
Resolution plan: .../resolution-plan.txt

Failures and missing evidence

None. All script steps succeeded; no credentials, deployment, or mapping evidence missing.

Agent triage by Codex, 2026-10-05. This is an agent assessment.

Sentry evidence now belongs to the daily check-in. The runbook removes the separate Sentry routine. Current D1 escalation and monitoring remain in #393. Closing this retired log does not resolve any Sentry issue.

sentry-checkin: run log for skilld-dev/skilld.dev #89

Description

Activity

harlan-github-agent commented on Aug 28, 2026

harlan-github-agent commented on Aug 28, 2026

harlan-github-agent commented on Aug 29, 2026

harlan-github-agent commented on Aug 29, 2026

harlan-github-agent commented on Sep 1, 2026

harlan-github-agent commented on Sep 2, 2026

harlan-github-agent commented on Sep 3, 2026

Sentry check-in — skilld.dev (project skilld) 2026-09-04, read-only

Scope and mapping

Deploy-state finding

Triage, one row per frozen id

Proposed actions

harlan-github-agent commented on Sep 4, 2026

Sentry check-in — skilld.dev (read-only)

Scope

Findings by cluster

Proposed actions

harlan-github-agent commented on Sep 6, 2026

harlan-github-agent commented on Sep 6, 2026

Sentry check-in: skilld-dev/skilld.dev (read-only)

Coverage (34/34)

Proposed actions

harlan-github-agent commented on Sep 7, 2026

Sentry check-in: skilld.dev (project skilld)

Scope and method

Triage, all 34 issues

Proposed actions

harlan-github-agent commented on Sep 8, 2026

Sentry check-in, skilld.dev, 2026-09-08 (read-only)

Scope and method

Dispositions

Proposed actions

harlan-github-agent commented on Sep 9, 2026

Sentry check-in, skilld-dev/skilld.dev, 2026-09-09T20:20:00.000Z

Scope and access

Counts

Deployment evidence

Resolved this run (9), all --in-release 103991d0b34830f98e5b361c4c8665d8f0d75aaa

Left open (33), all third-party, never resolved

Candidates

Artifacts

harlan-github-agent commented on Sep 10, 2026

Sentry Check-in: skilld-dev/skilld.dev — 2026-09-10T20:20:00.000Z

Scope and mapping

Coverage

Resolved this run

Unresolved, with reasons

Code repair Candidates

Artifacts

harlan-github-agent commented on Sep 11, 2026

Sentry check-in: skilld-dev/skilld.dev, 2026-09-11

Scope

Coverage

Dispositions

Sentry resolutions

Code repairs

Artifacts

Script failures and missing evidence

harlan-github-agent commented on Sep 12, 2026

Sentry Check-in: skilld.dev (skilld) — 2026-09-12

Run identity

Coverage

Dispositions (all 32 rows)

Resolutions

Code repairs

Notes

Artifacts

harlan-github-agent commented on Sep 13, 2026

Sentry check-in: skilld.dev/skilld.dev (org harlan-zw, project skilld)

Scope and mapping

Counts

Issue inventory (all third-party)

Resolutions

Code repair Candidates

Artifacts

Failures and missing evidence

harlan-zw commented on Oct 5, 2026

Sentry check-in — skilld.dev (project `skilld`) 2026-09-04, read-only

Sentry check-in: skilld.dev (project `skilld`)

Resolved this run (9), all `--in-release 103991d0b34830f98e5b361c4c8665d8f0d75aaa`

Left open (33), all `third-party`, never resolved