<quick_start> Tools in call order:
- Inputs — vertical, the prospect (if named) and the calculator inputs still missing (rooms, equipment age, failed captures, downtime, tickets, staff; per vertical: events/yr, headcount, time-to-first-frame).
- Gate Q (prospect named) —
qualify_lead; excluded → stop coaching for this contact and say why;STALE AE LEAD→ surface to Tim; junior title → coach the thread-up goal. - Gate S — Epiphan Knowledge
search(thensearch_product_catalogif a SKU is in doubt) for every product fact a question will carry: Pearl-2 channel count, Pearl Nexus channel guidance, EC20 publish path, Epiphan Edge pricing. Cache for the session; quote only what came back. Nothing back → the question drops the number and is tagged[spec unverified]. - Pull the matching questions from
reference/discovery-by-input.md; per input pick one of the four layers (Killer Q / Label / Job / Insight). Swap any anchor in the bank for the Gate S value. - Gate B —
check_my_copyon each question card's strings (they are said aloud to prospects). - Deliver the question cards (format below), coach delivery with the Riff rules, name the
ae_handoffcondition. Sidecar + run-log line. </quick_start>
<house_rules> Customer-facing skill. Before output, run the gates in order (details: ~/.claude/skills/epiphan-house-rules-skill/reference/gates.md): S: search_product_catalog (Epiphan AI) or Epiphan Knowledge search before any product claim. B: check_my_copy on every customer-facing string. Q: qualify_lead right after the candidate pull, before tiering/enrichment/drafts. D: outbound text becomes a Gmail draft, never a send. Identifiers (owner IDs, portal, emails, seats) come from reference/identifiers.md in that skill — do not restate them here. On any gate failure follow reference/self-healing.md (retry, degrade with a flag, alert Tim, halt) and append a line to ~/.claude/skill-runs/.jsonl. </house_rules>
The principle (read this first)
- Concise & killer. One breath. If it needs a comma to survive, cut it.
- Easy to riff. A frame they can say five ways, not a line they recite.
- Spec-true. No number or capability is stated from memory, including the anchors printed in the question bank. Gate S at the start of every run returns the current value; the bank's anchors are a reading aid last verified on the date in its header, and the run-time lookup wins.
- Master the voice, not the script. The skill's job is to make the words bulletproof so the rep's only job is delivery.
The method: each input is a question
The calculator can't compute until the rep earns these answers. For every input, the bank gives four layers — pick the one that fits the moment:
| Layer | Source | What it does |
|---|---|---|
| Killer Q | Never Split the Difference | a calibrated, no-oriented question — short, opens the room |
| Label | Never Split the Difference | "It sounds like… / It seems like…" — names the pain, gets a "that's right" |
| Job | Jobs To Be Done | the outcome the buyer is really hiring for |
| Insight | Challenger | the reframe that teaches them something about their own cost |
Output contract: question cards
One card per missing calculator input, in call order, delivered in chat (or as a file if the rep asks). Then the handoff line.
INPUT: <rooms | equipment age | failed captures | downtime | tickets | staff | events/yr …>
Killer Q : <one breath, no-oriented>
Label : "It sounds like …"
Job : <outcome they're hiring for>
Insight : <Challenger reframe; any spec here is the Gate S value or [spec unverified]>
Riff : <one alternate phrasing>
HANDOFF: ae_handoff (Epiphan AI) when a Manager+ economic buyer is in play and ≥3 inputs are earned.
The flow (one call)
- Open with a label, not a pitch ("Sounds like the older rooms are the ones eating your week.").
- Earn the inputs with killer questions — rooms, equipment age, failed captures, downtime, tickets, staff (and per vertical: events/yr, headcount, time-to-first-frame).
- Teach with the cost-of-inaction number the calculator returns (Challenger reframe).
- Tease the cheaper path — "there may be a more affordable path; not every room records at once" — a starting point, not a quote. Never name the mechanism.
- Hand off with
ae_handoff(Epiphan AI MCP) once a Manager+ economic buyer is in play.
Riff rules (so it stays natural)
- Drop the number, then go quiet. Let the silence work.
- Mirror their last 1–3 words to keep them talking before the next question.
- One question at a time; never stack two.
- If they're junior (Coordinator/Technician), your real goal is the name of whoever signs — thread up.
Guardrail — what the prospect hears
This coaches the rep, but the question strings are said to buyers, so Gate B applies to them. In
anything the prospect hears, never name the AV-matrix mechanism (matrix/switchboard/routing
switch) or third-party brands — "your CMS / LMS," pooled = "a more affordable path, a starting
point, not a quote." The AE owns fit, pricing, POC. See epiphan-ai-mcp-guide-skill.
<success_criteria>
- Every missing calculator input has exactly one question card (Killer Q + Label / Job / Insight layers)
- Questions are one breath long and sayable five ways
- Every spec inside a card equals the Epiphan Knowledge
searchvalue fetched this run, or is tagged[spec unverified]; nothing quoted from memory or from the bank's printed anchors - Cards passed
check_my_copy; a named prospect passedqualify_lead - The flow ends with a Challenger reframe on the cost-of-inaction number and, when a Manager+ EB is in play, an
ae_handoff - No buyer-facing string names the AV-matrix mechanism or third-party brands
- Junior contact? The coached goal shifts to threading up to whoever signs
- Outcome sidecar and run-log line written </success_criteria>
<self_healing>
Failure = no cards produced; partial = a stage degraded (name it: S returned nothing for an anchor,
B skipped, Q unavailable for a named prospect). Ladder: retry once → degrade (card without the
number, flagged) → alert Tim in the output header → halt only if the vertical cannot be
identified. Full contract: ~/.claude/skills/epiphan-house-rules-skill/reference/self-healing.md.
</self_healing>
<outcome_sidecar>
At the end of every run write ~/.claude/skill-analytics/last-outcome-cost-discovery-coach-skill.json
and append one line to ~/.claude/skill-runs/cost-discovery-coach-skill.jsonl:
{"ts":"<UTC ISO8601>","skill":"cost-discovery-coach-skill","version":"<frontmatter version>","variant":"default",
"status":"success|partial|error","runtime_ms":0,
"metrics":{"vertical":"","inputs_coached":0,"cards_generated":0,"specs_verified":0,"specs_unverified":0,"handoff_recommended":false},
"error":null,"session_id":"<date>"}
status partial/error must name the failing stage in error. The SessionStart sweep ingests it.
</outcome_sidecar>