Activity

1. harlan-github-agent

   added

   routine:daily-checkin

   on Oct 8, 2026
2. harlan-github-agent commented on Oct 8, 2026

   

   harlan-github-agentbot

   on Oct 8, 2026 – with Harlan GitHub Agent

   ContributorAuthor

   More actions

   [ACTION NEEDED] Daily check-in: 2026-10-08

   2026-10-08T08:23:00.000Z — daily-checkin on skilld-dev/skilld.dev | 2 found | 2 new | 0 already known | 0 outside allowed scope | 0 over 5 files | 2 issues requested

   AMBER skilld.dev

   All checks ran with full coverage after yesterday's credential block; a 10:24 deploy introduced a 1,969-event 503 storm on skills-raw.

   Check	Result	Detail
   Front doors	Pass	HTTP 200
   Deploy and CI	Pass	b1ce9025 live; three workflows passed
   Protected health	Warn	48 sync-failed Skills; token expiry
   Sentry	Warn	75 unresolved; 2 new
   Database	Warn	2 rejected job records
   Workers	Warn	1 exception, 1 load shed

   Needs you

   1. Rotate the production GITHUB_TOKEN before ~2026-10-22; repos denying the read App stop syncing when it expires.
   2. Review the Cloudflare escalation draft for Escalate recurring D1 internal errors to Cloudflare #393; internal errors recurred through 10:40Z.
   3. Review PR fix(registry): complete large repository source refresh #575; it closes the truncated-tree family and pins posthog, the source of the new storm.

   Changed since yesterday

   • New: SKILLD-3W, 1,969 events since the fix(registry): preserve root Skill snapshots #573 deploy. Unpinned Skills (posthog's tree always truncates) answer 503 with retry-after: 30 forever; crawlers honour it and loop. Candidate filed.
   • New: SKILLD-3X, one cold-key D1 overload on skill-related. Known family, monitored.
   • New: SKILLD-32: D1 returns "Currently processing an import job." but the retry matcher expects "a long-running import", so reads never retry. Candidate filed.
   • Recovered: credential coverage ([BLOCKED] Daily check-in: 2026-10-07 #570's blocker) and ai-ready:cron (ok, was partial).
   • Improved: sync-failed Skills 95 → 48, failed jobs 29 → 2, internalError outcomes 17 → 0.

   Unchanged work (4)

   • D1 internal-error escalation (Escalate recurring D1 internal errors to Cloudflare #393), draft awaiting review
   • X spend policy (X discovery projects past the accepted $40 monthly spend line #179), $21.75/mo estimate, below the $40 line
   • SKILLD-3V network-loss recurrence, 2 events, monitoring
   • Garbage-owner 404 writes, still in code, no fresh impact

   Evidence

   Exit 1 | coverage complete | deployment b1ce9025 | baseline 5.1h fresh
   Archive: ~/.local/state/daily-checkin/skilld-dev/skilld.dev/2026-10-08T14-04-21-207Z-4fe87104-db18-4736-a05a-8e61193f7fcb.md
   Pulse: 44 users, 20,398 Skills, 4 command copies (2 run, 2 install), 0 digests, $0 recorded AI cost, X $21.75/mo Post-only (74/2000 reads today); actual billing unavailable. Migrations match at 0148; no stale tasks.

   D1 returns "Currently processing an import job." but RETRYABLE_D1_READ_MESSAGES only matches "currently processing a long-running import", so the read retry never fires and the rejection 500s the request (Sentry SKILLD-32, 126 events since 2026-09-30, still recurring).

   Target: server/utils/db.ts

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

   Estimated to change 2 files.

   ---

   Since #573, skills-raw and skill-asset answer 503 with retry-after: 30 whenever rendered_commit_sha is NULL, and repos whose sync can never complete (posthog's permanently truncated tree) stay NULL forever, so every referenced-file request retried (Sentry SKILLD-3W, 1,969 events from 10:27Z).

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

   Verify with: curl -s -o /dev/null -w '%{http_code}\n' 'https://skilld.dev/api/skills-raw/posthog/posthog/diagnosing-experiment-results/references/qualitative-feedback.md'

   Estimated to change 3 files.