Best for
- The task you re-explain to the model every single time
- Teams where everyone prompts the same job differently, with different results
- Building a prompt library that raises the floor for everyone
What you give it
- The recurring task, your best and worst attempts at prompting it, and examples of good output
What you get back
- The template: fixed frame and variable slots separated — the engineering in the frame, the per-use work reduced to filling slots
- Slot specifications: per slot, what goes in, its format, its constraints, and a filled example — so misuse is hard
- The usage sheet: two worked examples end to end, the known failure modes, and when NOT to use this template
How it works
- Separates what varies per use from what should never vary: the stable frame absorbs all the engineering — role, rules, output contract, quality bars — so per-use effort drops to slot-filling.
- Builds the frame from the best existing attempts, with the mechanics applied: rules placed where they bind, output format explicit, unknowns handled, input slots fenced as data.
- Specifies each slot like an API parameter: content, format, constraints, a filled example, and what happens when it is empty or oversized.
- Ships with usage apparatus: worked examples, known failure modes, and the template's honest boundary — the tasks it is wrong for.
Example
You: We write competitor-update briefs weekly, and everyone prompts it differently with wildly uneven results.
Result: The template built from the team's best attempts: the frame carries the engineering (the analyst role, the sourcing rules — claims dated and attributed, the sceptical-of-vendor-claims stance, the fixed output sections, the 'state when evidence is thin' rule), and three slots carry the week (the raw material, the focus competitor, the deadline-driven length). Slot specs stop the known misuses (pasting unsourced rumour into the material slot gets caught by the sourcing rule). Two worked examples attached, plus the boundary note: 'not for deep-dive analyses — this template optimises for weekly cadence'. The team's output converged on the best member's quality within two weeks.
Limits — please read
- Templates serve recurring shapes; one-off tasks deserve one-off prompts — forcing the template onto edge cases produces confident mediocrity.
- The frame is engineered against today's cases; drift in the task erodes it, so the template carries a version and an owner.
- Quality ceilings are the frame's ceilings; a case-set evaluation (separate pass) is how upgrades get proven.