Why team-specific prompts beat general ones

GENERAL PROMPT                       TEAM-SPECIFIC PROMPT
──────────────────────────────────────────────────────────────────
"show cost"                           "payment-team's variable cost
                                       this week vs same week last
                                       month"

Returns: org-wide cost                Returns: payment-team's number
                                       (pre-filtered to relevance)

Generic, requires manual filtering    Pre-filtered to relevance

Slow for the asker                    Fast: direct to answer

Needs follow-up questions             Answers in one shot

1. SCOPE
   Which resources / accounts / tags apply
   Example: "team=payment-team, accounts=prod-us, prod-eu"

2. EXCLUSIONS
   Known things to exclude
   Example: "exclude ml-training-* resources (billed to data team)"

3. TIME WINDOW
   Specific period, not vague
   Example: "last 7 days vs same 7 days previous month"

4. OUTPUT FORMAT
   What the team wants to see (table, JSON, narrative)
   Example: "table with columns: resource_id, name, this_week,
            last_week, delta_pct, with a total row"

5. CONTEXT
   What makes this team's view different
   Example: "Highlight resources where delta_pct > 20%
            (significant for this team's volatility)"

PAYMENT-TEAM'S WEEKLY COST REVIEW PROMPT (saved as a skill):

"Show me the payment-team's variable cost for the last 7 days,
excluding ml-training-* tagged resources (those are billed
separately under the data team).
Compare to the same 7-day period last week.
Highlight any resources with >20% deviation.
Format: table with columns: resource_id, name, this_week,
last_week, delta_pct.
Include a row at the bottom for the team total.
Use this for our Friday cost review."

When invoked: agent returns the formatted table immediately.
Team's weekly review starts here, not with "what should we look at?"

GENERIC PROMPT TEST:
  Could a stranger use this prompt and get the right answer?
  → YES = generic

TEAM-SPECIFIC TEST:
  Does this prompt know our team's exclusions, conventions,
  preferred format?
  → YES = team-specific

Most useful prompts FAIL the generic test deliberately.

USE CASE                                  AGENT WORKLOAD WITHOUT
                                          TEAM PROMPT
──────────────────────────────────────────────────────────────────
Weekly cost review                        10 minutes back-and-forth
                                          (vs 30 seconds with prompt)

Anomaly investigation                      5 follow-up questions
                                          (vs immediate diagnosis)

Drafting a chargeback explanation         3 clarifications about
                                          allocation method
                                          (vs immediate draft)

Quarterly trend summary                    Negotiate scope each time
                                          (vs consistent format)

GOOD team-specific (parametrized):
  "Show team={{team}}'s cost for last {{days}} days,
   excluding {{excluded_tags}}"
  → Reusable across teams; configurable

BAD team-specific (hardcoded customer data):
  "Show team-acme-payments' cost..."
  → Locked to one customer; not portable; embeds secrets

Parametrize. Don't hardcode.

ANTI-PATTERN                              FIX
──────────────────────────────────────────────────────────────────
Too vague                                 Add scope + format
"show me cost stuff"                      "show payment-team cost
                                          last 7 days as table"

Too many options                          Pick one variant per skill
"could be table or chart or memo..."      "as a markdown table"

No exclusions                             Add known excludes
                                          (ml-training, etc.)

No success criteria                       Specify what "good" looks
                                          like (top 3 by impact)

Ambiguous time window                     "Last 7 days" or
"recently"                                "current quarter"

TYPICAL team prompt library:
  /payment-team-weekly          Weekly cost summary
  /payment-team-monthly         Monthly trend memo
  /payment-team-budget-status   Budget vs actual snapshot
  /payment-team-recs            Open recommendations digest
  /payment-team-anomaly         Anomaly investigation
  /payment-team-postmortem      Postmortem skeleton with cost

8-12 prompts cover most of a team's recurring cost work.

APPROACH A: generic prompt (10-minute review):
  T+0      Engineer: "show me last week's cost"
  T+5s     Agent: shows org-wide numbers
  T+30s    Engineer: "just my team please"
  T+45s    Agent: shows team but no comparison
  T+60s    Engineer: "compare to last week"
  ... 10 minutes of negotiation later ...
  T+10min  Engineer has the answer

APPROACH B: team-specific prompt as a skill (30-second review):
  T+0      Engineer invokes /payment-team-weekly
  T+30s    Agent: returns pre-filtered formatted table
            (scope, exclusions, format all in the prompt)
  T+30s    Engineer reads, identifies 5 anomalies to follow up
  T+30s    Total time: 30 seconds

10 minutes vs 30 seconds. Team prompts are the leverage.

For a team running 50+ cost reviews per year, the savings compound:
  Without team prompts: ~10 hours/year per engineer
  With team prompts: ~30 minutes/year per engineer

QUESTION 1: __________
  Generic version: "__________"
  Team-specific version:
    SCOPE:        __________
    EXCLUSIONS:   __________
    TIME WINDOW:  __________
    OUTPUT:       __________
    CONTEXT:      __________

QUESTION 2: __________
  (same template)

QUESTION 3: __________
  (same template)

NEXT STEP:
  □ Save the most-recurring one as a skill (M6.4.L2)
  □ Share with the team (M6.4.L3)
  □ Schedule quarterly review (M6.4.L4)