메뉴
Triage and respond to the user-feedback queue — pull open feedback, detect items already fixed but never marked, route translation-requests and metadata-corrections, mark items resolved, and (on approval) email submitters. Use when asked to check/triage/answer/clear feedback, "what are people saying?", or "respond to feedback."
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Run OCR and translation quality evaluations across scripts and languages. Produces research-grade reports with MCR, cross-model agreement, embedding-space hallucination detection, and corpus readiness scores.
Run a Glass Bead Game session — a parlor game played against the Source Library corpus. Players connect verified passages from different books, times, and traditions into a growing string of 'beads,' with Claude as librarian, verifier, and scribe. Trigger when the user says 'glass bead game,' 'let's play beads,' or asks to start/resume/publish a game session. Claude runs the search-verify-display loop; humans do the noticing, choosing, and judging.
Stage-2 verification of first-translation flips using INDEPENDENT Claude subagents (subscription, not API key) with full evidence capture. Use to verify proposed badge changes (demotes/promotes) before writing any public claim — the binding step that catches AI-fabricated priors. Each subagent reports its actual search queries + sources, persisted to the first_translation_attempts provenance log.
Sweep Claude's private per-machine memory for team-relevant operational knowledge and propose moving it into the shared repo (or the private ops repo) as draft PRs. Use to promote lessons that accreted privately, or run on a schedule. Never auto-merges; never leaks.
One-shot audience & usage report — signups, DAU/MAU, time on-site, traffic, top content, search behavior, referrers, countries, and user journeys. Use when asked "what's our usage?", "how many signups?", "time on site?", "traffic?", or any audience/engagement metrics question.
Use this skill whenever choosing or evaluating a background image for the 'quote' section of a Source Library collection page — the wide banner where a sourced quotation is overlaid on an image drawn from the collection's plates or illustrations. Trigger for any request to pick, rank, or judge a background image behind quote text, select a plate for the quote banner, or decide whether a candidate image works as a text background. The goal is a beautiful, legible, on-theme backdrop that the quotation reads cleanly over.
| name | feedback |
| description | Triage and respond to the user-feedback queue — pull open feedback, detect items already fixed but never marked, route translation-requests and metadata-corrections, mark items resolved, and (on approval) email submitters. Use when asked to check/triage/answer/clear feedback, "what are people saying?", or "respond to feedback." |
The feedback collection (Mongo bookstore) collects messages from the
FeedbackWidget, the translation-request prompt, and the MCP submit_feedback
tool. The infra to resolve it already exists — this skill runs the loop instead
of the throwaway _tmp_fb_* scripts that used to do it ad-hoc.
read != trueread == true && addressed != true (triaged, not resolved)addressed == true (done)Resolving via the admin API (PATCH /api/feedback/[id]) auto-emails the
submitter once if they left an email (feedback-reply-email.ts, idempotent via
reply_sent_at). The /admin/feedback page is the human UI for this.
set -a; source .env.production.local; set +a
node scripts/analytics/feedback-triage.mjs # all open items, grouped
node scripts/analytics/feedback-triage.mjs --days 45 # recent only
Groups: bug · translation-request · metadata-correction · partner-cms ·
feature-request · praise · other. Each line shows the _id, date, state,
whether an email was left (✉), and the page. The footer splits emailed (reply
candidates) vs anonymous (bulk-resolvable).
For each open item, decide if it's already shipped but never marked — this is the main backlog source. Cross-check, don't guess:
git log --oneline --since=<feedback date> -S "<keyword>" and search merged PRs (gh pr list --search).memory/ + MEMORY.md for a matching lesson (e.g. librarian "kites" → PR #2829; Spanish voice → PR #2699).Propose a disposition per item: already-fixed (link PR/page) · real, route it · wontfix/explain · praise (ack).
book_id/page and hand to the daily pipeline queue (the corpus is paused as of 2026-06-08 — note that; don't unpause without Derek's say-so). Mark addressed once queued./catalog/*/edit notes) → one GitHub issue for the BPH catalog editor; tag EFM-facing.When you file a GitHub issue from a feedback item, LINK it back so the loop closes itself. Put a marker line anywhere in the issue body:
Feedback-ID: <mongo _id> # e.g. Feedback-ID: 6a473930a5f38bf2d79673bf
One issue can carry several ids (comma/space separated, or repeat the line).
Then when that issue closes (a merged Fixes #N PR closes it automatically),
feedback-reconcile-issues.mjs (Step 5) marks the linked feedback addressed
with a link to the issue — no manual re-triage. Only Fixes #N an issue when
the PR resolves the WHOLE issue. Partial fixes → reference it (Refs #N) and
leave it open, or split the issue, or the reconciler will prematurely resolve
feedback that isn't actually done (this bit us on #3051 — artwork half shipped,
thumbnail half didn't, and Fixes #3051 closed the lot).
Anonymous / no-email items (the bulk): mark directly — no email is sent.
node scripts/analytics/feedback-triage.mjs --mark-addressed <id,id,...> --note "what was done" [--link URL] --apply
# or just triage without resolving:
node scripts/analytics/feedback-triage.mjs --mark-read <id,id,...> --apply
Always dry-run first (omit --apply). The note becomes addressed_action.
Items WITH an email that deserve a reply: these must go through the
/admin/feedback UI (or the authed PATCH /api/feedback/[id]) so the canonical
reply email fires. Script-marking sets state but does NOT email — the script
warns when a marked id has an email. Sending an email is outward-facing and
unsendable: draft it, show Derek, and only resolve-with-reply on his explicit
approval. Per-item, not blanket.
The main backlog source is feedback that was fixed but never marked — often
via an issue+PR in another session. Rather than re-verify by hand every time (a
naive keyword→PR-title matcher is useless — measured 99/104 open items falsely
"matched" on common words), close the loop off explicit Feedback-ID: links
(Step 3):
node scripts/analytics/feedback-reconcile-issues.mjs # dry-run
node scripts/analytics/feedback-reconcile-issues.mjs --apply # mark linked feedback for CLOSED issues
node scripts/analytics/feedback-reconcile-issues.mjs --state all # also list open-issue links (informational)
It scans user-feedback issues, reads their Feedback-ID: markers, and marks
the linked feedback addressed only when the issue is closed (idempotent —
skips already-addressed rows; writes Mongo state only, never emails; warns on
rows with an email so a human can send the reply). Run it at the top of every
triage pass so resolved items drop off before you look at the queue.
_id list, after classifying. A wrong "addressed" can fire a wrong email.172.x), so logged country ≠ real location. Don't infer audience geography from feedback rows.system_config.processing_control.paused) — translation-requests queue but won't process until unpaused.