ZenUml/conf-app / .claude/skills/customer-followup
customer-followup
VerifiedPrioritise Confluence customers for conversion, retention, win-back or pause using lifecycle, feedback, license, usage and conversation evidence. Draft follow-ups and update the tracker within the user's authorised scope; reuse existing query skills.
Customer follow-up
Turn verified customer facts into one useful next step. Keep business decisions independent of the storage tool.
Start with the current task
Read shared evidence rules, contact and action policy, and record contract. For a batch, use 1by1 for review pace; existing explicit authorisation remains valid.
- Identify the requested customer(s), product(s), purpose, time window and already authorised actions from this conversation. Do not ask the user to repeat them. For a general recent scan, use the last seven complete days and state the timezone; expand to 30 days when useful and label the change.
- Read the existing opportunity and activity records first. Check outstanding drafts, scheduled messages, promises, pause conditions and previous declines before preparing another contact. History search supplies context; original systems establish current state.
- Fetch only evidence needed for the decision:
- Recent changes:
customer-lifecycle. - User opinions, problems, requests:
customer-feedback. - Site/product/license/payment/contact:
tenantandmarketplace. - Engagement: use
conf-appfor both Mixpanel and D1; complete the shared evidence coverage gate before declaring inactivity or skipping follow-up.client-healthis a Lite-only relative shortlist, not conversion probability. - Space inventory:
macro-count, with its coverage limitations in the shared rules. - Communication: Gmail, support tickets and
search-conversation-history.
- Recent changes:
- Resolve contact role separately from site identity. A Marketplace technical contact can be an employee, a partner, an ex-employee or an unknown role. Never label them site admin without evidence.
General scan with no explicit signal
When no customer is named and customer-lifecycle and customer-feedback return nothing that needs a reply, do not report "nothing to follow up". Rank by behaviour instead:
T=$(mktemp -d)
python3 .claude/skills/client-health/scripts/health_score.py --json --top 60 > "$T/health.json"
python3 .claude/skills/customer-followup/scripts/behaviour_targets.py --health "$T/health.json" --exclude "$T/exclude.json" --top 10
python3 .claude/skills/marketplace/scripts/mp_report.py whois <domain> --json # per candidate, live: --local skips the Lite space-licence (KV) layer
Build exclude.json ({"<bare subdomain>": "<reason>"}) from the tracker's paused and recently contacted opportunities in $T or private/; never commit it. If the tracker cannot be read, omit --exclude and label the result "tracker not checked". For each candidate read the tracker stage and the last activity's date, channel and outcome, then return it as one of: never contacted, contacted, no reply (date, channel), paused (resume condition) or in conversation. When the tracker shows earlier outreach, state the reply rate per channel and choose the next channel by what got replies. A COMPED space licence is a support grant: find its ticket and count it as contact. skip rows already pay for another app; unknown rows have no Lite licence in the snapshot. Apply the contact policy unchanged: partner rows (partner_name) and unverified technical contacts follow its responsibility rules, and no ranking or telemetry goes into a customer message.
Decide, with evidence
Give the customer name, goal, strongest signal, contactability, blocker, one recommended next step, and any material unknown. Prioritise concrete demand, a recent reply or referral, verified usage and a reachable appropriate person. Potential annual value helps prioritise; seats alone do not prove usage or willingness to pay.
| Direction | Use when | First useful next step |
|---|---|---|
| Conversion | There is an actual need and a plausible purchase route | Clarify one buying condition or offer a forwardable Full quote |
| Retention | Recent installation/use, limited paid value, or experience is uncertain | Ask the main user about their workflow and experience |
| Win-back | Explicit unmet need, cancellation or verified uninstall | Understand the gap and whether a relevant fix merits another try |
| Pause | No reciprocal contact, explicit delay/decline, or our promised feature is unfinished | Record a concrete resume condition; prioritise other customers |
A Full trial approaching expiry is an observation, not urgency or proof of failed conversion. Verify billing/renewal evidence and actual blockers before proposing contact. A payment, a commercial license, a support extension and a communication reply are different facts.
For a purchase/renewal opportunity, a matched valid non-zero order plus an effective commercial license establishes the commercial result: mark won / 已成交 even when the transaction's payment status is Open. Record settlement separately. Exclude refund, cancelled/voided and free-only transactions; a license alone is insufficient. Apply the transaction rules in shared evidence, including source identifiers and overlapping-order uncertainty.
On an authorised status-only refresh, reconcile the existing stage, blocker and next-action text with new facts. An established purchase must not remain “procurement / awaiting approval” solely because settlement is unknown or the last email predates the order. Update those stale statements and append the evidence; do not invent outreach or a new follow-up plan. Preserve user-directed pauses/observation scope and unresolved technical work separately from the commercial result.
Reconsider paused customers only when a subsequent authorised run finds the recorded condition satisfied; record why. This skill does not create background monitoring.
Prepare and perform the next step
Apply the contact policy. Use a short natural message with one easy reply, verified product name and relevant context. Draft when drafting is requested. If the user already authorised the actual send/schedule/grant or record update, complete it within that scope without another confirmation. Otherwise prepare a concrete reviewable result before asking for any necessary authorisation.
For a ZenUML Lite paid-upgrade email, use zenuml-upgrade-email to verify the claims, write or revise the message, and save its Gmail draft. This flow still owns customer selection, commercial direction, authorisation and the tracker handoff.
Before sending, re-check recipients, latest replies, equivalent drafts/scheduled messages and the final content against scope. For scheduling, use the recipient's verified timezone, explicit calendar date and daylight-saving rules; verify what the mail system actually scheduled. A spreadsheet date is not a scheduled email.
Verify the result in the original system. An attempted or ambiguous send is 待核验, not 已发送; inspect before retrying to prevent duplicates. Then append one activity record and update the opportunity summary. Never silently roll back an external action because a tracker write failed: report the discrepancy and reconcile from evidence.
Storage
Use customer-tracker for tracker writes and saved-result verification. Pass the authorised scope, stable opportunity IDs, verified facts/decisions, original source timestamps and links, and any missing evidence. This flow owns investigation and business decisions; customer-tracker owns applying them to the workbook without starting another investigation.
The record contract remains authoritative for record meanings; private/operations/customer-followup-tracker.md maps them to the current workbook. Keep real customer data and investigation archives in private/. Do not put tracker IDs, contacts or customer examples in public skill files.
If there is no storage connection, finish the investigation/draft and prepare an exact pending record privately; clearly distinguish it from a saved update. Do not claim the tracker changed until re-read verification succeeds.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |