Deal Red Flags

Deal Red Flags

Review the written record of a single deal - call notes, opportunity-record text fields, CRM activity log entries, email threads - and surface qualification red flags with the evidence behind each one. Free-text notes are lossy, so silence in the notes is not the same as a confirmed problem.

Grade every flag on two independent axes, never collapsed into a single score:

• Severity: how much damage the risk does to this deal if it is real.
• Confidence: how the notes evidence it.

A high-severity but poorly evidenced gap is escalated differently from a low-severity but certain one. The most common and most useful verdict is Unknown - the notes say nothing either way - because it converts a note gap into a question for the next call. A structured scorecard cannot produce Unknown honestly, since its input is already structured; that asymmetry is this skill's whole reason to exist.

Hard rules, not suggestions:

• Every finding graded Verified or Assumed must quote the exact fragment of the notes it came from. A finding with no quotable evidence is by definition Unknown.
• Every flag has a specific piece of disconfirming evidence that clears it. Check for it before recording a risk, so the review is not a one-way ratchet where every deal looks doomed.

Use this, or use a sibling

Use this skill when the user has the written notes of one deal and wants its risks surfaced from the prose. Use something else when:

• The user wants a structured score against MEDDPICC criteria, criterion by criterion - mbfinotti/sales-skills@meddpicc-scorecard. That skill scores structured answers; this one reads messy prose and reports what the prose cannot support. This skill never outputs a methodology score.
• The user wants the stakeholder map built (who is the champion, economic buyer, blocker) - mbfinotti/sales-skills@deal-champion-mapping. This skill only flags that champion evidence is missing or weak; it never names likely personas or builds the map.
• The user wants a full discovery question set - mbfinotti/sales-skills@sales-discovery-questions. This skill emits exactly one diagnostic question per gap.
• The user wants the ROI or business-case narrative - mbfinotti/sales-skills@deal-value-calc.
• The user wants these same notes turned into a follow-up email - mbfinotti/sales-skills@sales-meeting-recap.
• The user has a pipeline export of many deals. That is pipeline-level work (hygiene, forecasting) and out of scope here: this skill reads the written record of one deal, deeply.

Interview

Ask before reviewing - one question per message, multiple-choice where possible. Skip anything the notes already answer.

• B2B or B2C? (B2C here means considered purchases with a real sales conversation - remodeling, financial products, high-ticket coaching - not impulse retail.)
• Rough deal size and expected cycle length? (Calibrates severity: a missing mutual action plan is severe on a six-month enterprise deal, irrelevant on a two-week transactional one.)
• What stage does the seller believe the deal is in; is it on a forecast or commit list, and by what date must the answers land - the forecast call, the stage gate, or the buyer's own event? (A committed deal raises the stakes of every gap; a gap that resolves after that date is not worth chasing at any price.)
• What period do the notes cover, and are they complete? (One call's notes cannot evidence trends; see failure modes.)
• Is the goal this deal, or the habit behind it? (Saving this cycle ranks buyer-facing asks first; fixing qualification or note hygiene promotes the record work that pays across every later deal.)
• What is the effort ceiling before the next gate - how many asks the relationship can carry, and whether a manager or exec sponsor is available? (Decides which chase and response rungs are open at all.)
• Anything the user already suspects? (Becomes a hypothesis to test against the evidence - never a conclusion to confirm.)

Intake

Accept any prose about the deal:

• Raw call notes.
• A pasted activity log.
• Opportunity-record free-text fields.
• Email or chat threads.
• Recap emails.
• Transcript excerpts.

Notes exported from any CRM or conversation-recording tool work as input - nothing in the workflow depends on which tool produced them. If your environment can read files, accept a file path or export; otherwise ask the user to paste the text.

Partial, messy, or thin notes are fine - saying what is missing is half of this skill's job. Be explicit with the user about what the review can and cannot see: it grades the notes, not the deal, so thin notes produce many Unknowns, and that is itself a finding about deal hygiene.

Workflow

1. Run the Interview; then confirm the input covers everything the user has (a stray email thread often holds the only first-hand buyer quote).
2. Normalize the notes into a dated timeline of events: meetings held, who attended, commitments made, dates moved. Note where the record thins out - that is evidence too.
3. Extract candidate evidence: verbatim fragments that assert or disconfirm anything in the catalog. Distinguish first-hand buyer statements from the rep's own inferences and from second-hand relays - "I have it in writing from the economic buyer" outranks "my champion told me".
4. Run every group of references/red-flag-catalog.md against the timeline and the evidence. Resolve every flag in every group to a state - never skip a group.
5. Check each flag's disconfirming evidence before recording a risk. A flag whose clearing evidence is quoted in the notes is Cleared, not Red.
6. Assign severity to every non-Cleared flag, calibrated to the deal size, cycle length, and motion from the Interview - not to a fixed table.
7. Separate what is evidenced from what is merely absent. Write exactly one diagnostic question per Unknown flag (the catalog supplies one per flag), then order the Gaps by the chase order below - the rep gets one next touch, not twelve.
8. Write the report in the output shape below, attaching one remediation play per Red and Amber flag from the catalog and naming the response rung that play belongs to.
9. Run the Quality gate; iterate - re-grade and rewrite - until every item passes.
10. If your harness has persistent memory, record the flags raised, their states, and their evidence, so the next review of the same deal can check which gaps closed. Otherwise end the report with a short carry-forward list for the user to keep.

Grading rubric

Confidence states - how the notes evidence the flag:

Severity - High, Medium, or Low: the damage to this deal if the risk is real, judged against deal size and cycle length, not a fixed table. Grading a flag red/amber/green with a verified-versus-assumed confidence test is standard practice in deal inspection (documented across MEDDPICC-lineage sources); the state names above extend that vocabulary with Unknown for note review.