Skill Details
Runtime
(claude-4-6-sonnet)
Example user message
[userMessage]: i want a prompt that turns messy meeting notes into clean action items
Example agent response
CLARITY READ: GOAL clarity = HIGH (clear job: notes -> action items). Biggest gap: [OUTPUT_FORMAT] and whether owners/dates exist in the notes - bracketed.
SYSTEM PROMPT:
You convert raw meeting notes into a clean action-item list. Extract only what was actually said. Never invent owners, dates, or decisions. If an owner is missing write [unassigned]; if a due date is missing write [no date]. Be concise and literal.
USER PROMPT TEMPLATE:
Notes:
"""
[MEETING_NOTES]
"""
Task: Extract every action item. For each row output: Owner | Action | Due | Priority. Set Priority only from the language used (LOW/MED/HIGH); if unclear, MED.
Output format: a markdown table with those four columns. After the table, add an "Open questions" list for anything ambiguous.
Self-check before answering: (1) Did I invent any owner or date not in the notes? Remove it. (2) Is every row traceable to a line in the notes? (3) Are priorities bands only, no scores? Revise once, then output only the final table + open questions.
VARIABLE KEY:
[MEETING_NOTES] = paste the raw notes, e.g. "Sarah sends the deck Fri, login bug needs fixing ASAP, revisit pricing next week."
TUNING NOTES:
- For JSON instead of a table, swap the format line for: output a JSON array of {owner, action, due, priority}.
- Claude/GPT/Gemini: portable as-is.
NOT a transcription tool and NOT a decision-maker - confirm owners and dates with attendees before acting.
By purchasing this skill, you agree to our terms of service
CLAUDE-4-6-SONNET
Turn a rough goal into a reusable, model-agnostic prompt SYSTEM - not a one-off.
✅ System prompt (role, guardrails, refusal rules)
✅ User template with [VARIABLES] you reuse every run
✅ Variable key + sample values
✅ Self-check loop (model critiques + revises before answering)
✅ Tuning notes for Claude, GPT, Gemini & open-source
Honest by design: architects prompt TEXT - never fakes a score or invents facts (unknowns become [brackets]). Rates input clarity LOW/MED/HIGH. For prompt engineers.
...more
Added 1 week ago
