ZenUml/conf-app / .claude/skills/customer-feedback
customer-feedback
VerifiedFind and reconcile explicit customer feedback across Marketplace cancellation reasons, Mixpanel satisfaction submissions, in-product D1 feature/problem reports, in-product Fullscreen recovery requests and support conversations. Use for recent feedback, unmet needs, feature requests, trial or churn investigations and product research. Keeps user opinions separate from behavioural evidence and preserves source coverage.
Customer feedback
Read shared evidence and source rules. Scope by site/product, date window and question. Seven complete days is only a shortcut for "recent"; use the requested 30/90-day period or incident window for research. This is a read-only investigation, not permission to contact customers.
Run to completion — do not stop to ask
Default scope. When the invocation gives no site, product, window or question, run immediately with: all products, the last seven complete days (state the dates and timezone in the report), question = "what explicit feedback arrived". Do not ask for scope; a missing argument is not a blocker. Ask only when the request is contradictory or the scope cannot be defaulted.
Completeness gate. All five sources in the table below must be queried before the report is written. Every source ends in ok, empty, failed or unavailable. not_queried is allowed only after a real attempt was blocked (credential rejected, endpoint error after one retry); name the blocker. "I stopped to check with you" and "ran out of steps" are not blockers.
Paginate to the end. A Gmail search, JSM search or any paged list counts as queried only after the last page is read (no nextPageToken). A first page is partial, and partial is not a final state.
No mid-run options. Every source here is read-only and needs no approval. Do not offer numbered choices, ask "should I query X", or hand back an interim report before the gate is met. Present options only after the full report, and only for real follow-up decisions.
Self-check before Return. Walk the Sources table row by row and confirm each row's status, record count and window. Gmail and the JSM ZEN queue are two separate checks inside the last row; confirm both. For Fullscreen recovery, confirm the request count and that each listed request has its status result and age. If any row or check is not_queried or partial, go back and finish it.
Sources
| Source | What to retrieve | Boundary |
|---|---|---|
| Marketplace | Cancellation/unsubscribe reason, customer free text, product/subscription and feedback date | State change alone has no customer explanation; cancellation is not necessarily uninstall |
| Mixpanel | Actual satisfaction submission, option/score/text and identity fields supported by current event catalog | Opens/clicks/errors describe behaviour, not an expressed opinion; submission workflow is not another feedback item |
| D1 in-product feedback | Feature requests, problems, description, entry surface, product/site/user context, report reference and optional attachments | Not limited to satisfaction; verify table, deployment and coverage before querying |
| Fullscreen recovery requests (Lite) | Pending requests: node scripts/fullscreen-recovery-support.mjs list, then status --id <uuid> per request (runbook). Report request id, site (map cloudId through the Marketplace snapshot), age from createdAt, scenarios and mail state | list returns only requests with no recorded support outcome; a request with no sent reply is owed one: report its age. Needs FULLSCREEN_RECOVERY_API_URL (production https://conf-lite.zenuml.com) and FULLSCREEN_RECOVERY_OPERATOR_TOKEN (one secret from the backend's FULLSCREEN_RECOVERY_SUPPORT_TOKENS); missing locally (CLI prints Invalid FULLSCREEN_RECOVERY_API_URL or FULLSCREEN_RECOVERY_OPERATOR_TOKEN is required) or Backend status 503 = unavailable, Backend status 401 = failed, {"cases":[]} = empty, 50 rows = ok (at list cap): the list has no next page, so say newer requests may be missing. Decrypted email and note stay private |
| Gmail/support tickets | Relevant messages, actual sender, quoted need, replies, resolution and promises. Query both the support mailbox (every result page) and the JSM ZEN queue (support-queue skill, local agent token) | Prior agent summaries are search clues; read the original thread before claiming current outcome |
D1 capability check
Inspect current schema/migrations and deployed query access first. FeedbackReport work existing only in another checkout is not proof of production support. Use the configured read-only D1 query route and inspect table columns before selecting; do not invent column names or treat a missing table/failed query as no feedback. Record screenshot retention separately from report retention, and do not download or expose attachments unless needed for the investigation. Existing CSAT and feature feedback are distinct entry points.
Gather and reconcile
- Resolve the site and product from verified identifiers. Keep contact role uncertain where evidence is missing.
- Retrieve original feedback records within the chosen window and the surrounding thread or behavioural window necessary to interpret them. Use existing
marketplace,mixpanel,conf-appand conversation tools for access rather than new credentials or duplicate exports. - Preserve the user's words in a clearly labelled field; separately classify feature request, bug, unmet need, usability, satisfaction, procurement or other. Put interpretation in a different field.
- Link records only using a verified shared report/message reference. Same user/site/time is at most a suspected association. A D1 report plus a Mixpanel submit-success with no shared reference stays one explicit report and one possibly related process event; do not count the latter as a second opinion or silently merge it.
- Behaviour can suggest what to ask, not why a customer cancelled. For example, successful PNG download and failed email export are different actions; neither proves the reason for an unmet-needs response.
Handled items
Before writing the report, read private/operations/customer-feedback-handled.json. Leave out every record whose source ID is listed there, and say in one line how many were left out. A message, comment or report on the same thread or ticket that is newer than its handled_at is reported again. Gather sources as usual: the ledger filters the report and does not replace any query. When the owner says items are handled, add them to the ledger. For a Fullscreen request, also record its supportOutcome through the support CLI (outcome, dry run first), so that list stops returning it.
Return
Return the shared signal fields plus feedback category, original-text reference, resolution/reply evidence, and relationship to other records (verified, suspected, unrelated). Separate explicit feedback totals from behaviour/process events. Include per-source freshness and successful-empty / failed / unavailable / not-queried coverage. Do not summarise missing sources as "no feedback".
Use customer-followup when the user also wants prioritisation, communication or a tracker update; the same evidence should be reused without re-querying unchanged sources.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| finishing-a-development-branchobra/superpowersUse when implementation is complete, all tests pass, and you need to decide how to integrate the work | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| systematic-debuggingobra/superpowersUse when encountering any bug, test failure, or unexpected behavior, before proposing fixes | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| test-driven-developmentobra/superpowersUse when implementing any feature or bugfix, before writing implementation code | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| skill-creatoranthropics/skillsCreate new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |